Cataloguing the Eighteenth Century

Eighteenth Century Collections Online is based on a microfilm collection produced between 1982 and the early 2000s. This film collection itself was based on a catalogue of books begun in 1976 called the Eighteenth-Century Short Title Catalogue, under the editorship of Robin Alston, consultant at the British Library, co-edited with Henry Snyder in the United States. From 1987 this project expanded to include material printed before 1700 and was eventually renamed the English Short Title Catalogue (ESTC). In its current form the ESTC is an online catalogue of printed material published from the fifteenth century to the end of the eighteenth century. It’s difficult to capture the sheer scale of the ESTC and its ambition: looking back over the project from 2003, Thomas Tanselle reaches for numbers: the ‘file (achieved at a cost of about 30 million dollars) consists of some 435,000 records, indicating the location of over 2,000,000 copies in 1,600 libraries around the world’ (in Snyder & Reference SmithSmith, 2003: xi). Currently it comprises more than 480,000 records from 2,000 libraries worldwide.Footnote ² However, the project was initially confined to material printed between 1700 and 1800, and because it is this that shaped the underlying nature of ECCO, this history concentrates on the catalogue project before 1987.Footnote ³

The first discussions about the possibility of a catalogue that would cover the eighteenth century began as early as 1962 amongst members of the Bibliographical Society, and such a catalogue was perceived as the logical next step from the two Short Title Catalogues (STCs) covering material printed between 1475 and 1640 (Pollard and Redgrave) and 1641 and 1700 (Wing). However, it was from 1975 that the catalogue got the necessary backing to start. More significantly, the discussions and plans by the leading editors for the 18thC STC emphasised the necessity that it be produced as an electronic file so that it could be managed and queried by computers (Reference Alston and JanettaAlston & Janetta, 1978; Reference Crump, Snyder and SmithCrump, 2003: 54; Reference Snyder, Snyder and SmithSnyder, 2003: 21–30; Reference AlstonAlston, 2004). One of the most significant aspects of the 18thC STC is that it was deeply influenced by developments in computer-aided library cataloguing and online networked computer systems in the 1960s. The technology of machine-readable catalogue records would be an essential aspect of how metadata (that is, information about an object, distinct from the object itself) works alongside images and text of old books in digital archives and collections of the twenty-first century.

Korshin’s 1976 grant application to get the 18thC STC off the ground included the participation of Hank Epstein, the director of the Stanford University computing team (Reference AlstonAlston, 2004). Notably, the Stanford Research Institute was the first, in 1963, to demonstrate an ‘online bibliographical search system’, an ‘online full-text search system’, and systems that could be used remotely over long distances, and it was the first to use a screen display for interaction between human and computer (Reference Bourne and HahnBourne & Hahn, 2003: 14–15). In 1978 the Stanford-based Research Libraries Group developed an online networked database called the Research Libraries Information Network (RLIN).Footnote ⁴ In 1980 the 18thC STC at the British Library formed an important collaboration with the group, and by 1985 US and UK teams of the 18thC STC were able to edit the same file interactively online (Reference Crump, Snyder and SmithCrump, 2003: 55).

Technological solutions to managing information had been the subject of both visionary projects and practical application since the end of the nineteenth century; two particular figures are often cited as seminal thinkers in this field. One is Paul Otlet (1868–1944) who, with Henri La Fontaine, designed a huge card catalogue in the 1890s entitled the ‘Universal Bibliographic Repertory’. Otlet worked with Robert Goldschmidt on microfilming in the 1920s and 1930s, after which Otlet published a collection of his essays on the future of information science, Traité de Documentation, in 1934. The other is Vannevar Bush (1890–1974): in 1945 he proposed – but never built – a machine called ‘Memex’ that would display microfilm and that also had the capability to apply, search, and retrieve keywords about the information on the microfilms (Reference Deegan and SutherlandDeegan & Sutherland, 2009: 125–6; Reference BorsukBorsuk, 2018: 209–13).

However, more significant than either of these men for the future of library information was Henriette Avram (1919–2006). Her career as one of the first computer programmers led eventually to designing library information systems at the Library of Congress in 1965. It was here that she developed the first – and what would become the standard – computerised library cataloguing system throughout the world: MAchine Readable Catalogue (MARC) (Reference Rather and WigginsRather & Wiggins, 1989). Before the advent of computer-aided catalogues libraries used card catalogues; the details about a book, for instance, would be recorded on a three-by-five card. The system for ensuring all libraries had access to and could update library catalogue records involved card-copying services and transporting duplicate card records by mail. By contrast a record that a machine can read can be disseminated and centralised much more easily. A MARC record is divided up into coded fields each of which contains a designated type of information, such as author, title, library location, subject, and many more. Avram’s pioneering work necessitated a thorough understanding of computing and the principles of bibliographical cataloguing. The MAchine Readable Catalogue, then, was not only about designing a record to be parsed by a computer; it also set the standard for bibliographical records that libraries across the world would follow. The eventual result is the kind of human-readable record you can see on the online ESTC catalogue entry for the 1789 issue of Patrick Browne’s The Civil and Natural History of Jamaica (Figure 1). Later I use this book to explore the relationship between the physical book and its record.

Figure 1 ESTC record, Patrick Browne, The Civil and Natural History of Jamaica, 1789

(screenshot, 22 May 2020)

In fact, Alston discussed the 18thC STC with Henriette Avram in 1977 (Reference AlstonAlston, 2004). In the earliest discussions the catalogue was to adopt the principles of the seminal catalogue of early print: Pollard and Redgrave’s 1926 A Short-Title Catalogue of Books Printed in England, Scotland, & Ireland and of English Books Printed Abroad, 1475–1640. But it was clear that the 18thC STC would be a computerised catalogue and that therefore records would follow the much more detailed standards required by MARC (Reference Alston and JanettaAlston & Janetta, 1978: 24–6). In his 1981 lecture ‘Computers and Bibliography’ Alston was adamant that computing would enable more powerful scholarship. Alongside the inventions of writing and printing as technologies of knowledge, Alston noted, ‘we have now added a third (and by comparison with the former two) quite remarkable one: the storage, and virtually instant retrieval, of information about the present and the past in electro-mechanical form, and the mechanical aids available to assist in this process are now formidable – more powerful than anything we have known’ (Reference AlstonAlston, 1981a: 379).

Using machine-readable records meant that users could perform much more sophisticated searches than were possible with card catalogues (Reference ZeemanZeeman, 1980: 4). In addition, cataloguers, ‘keen to facilitate greater access to the online records, tended subconsciously (perhaps) to transcribe long titles’, enabling users to conduct complex keyword searches (Reference CrumpCrump, 1988: 5).

However, using MARC for the 18thC STC would not be a seamless fit. It’s worth reminding ourselves that each book in this period was handmade, the product of a series of processes which depended ‘upon a complex sequence of events, all of which were determined by humans capable of fallibility, stupidity, laziness, inconsistency, disobedience’ (Reference AlstonAlston, 1981a: 372). These included the making of paper, ink, and metal type, as well as composing type into sentences, locking those sentences into a frame, using the press to make each sheet, proofreading, and compiling it all into a book (Figures 2 & 3).

Figure 2 Metal type

Figure 3 Hand press, c.1700s

As Alston notes, ‘data must be stored within a system hospitable to eccentric evidence’ (Reference AlstonAlston, 1981a: 379). Books of the hand-press period and their records are ‘eccentric’ in the sense of being ‘Irregular, anomalous’ (OED, 6a). That is to say that – leaving aside editions and variants – the potential disparity between copies of the same book requires a method of cataloguing that can register such nonconformity. Alston emphasised the human agency involved in making books in order to highlight the limitations of the new ‘bibliographical networks’ that require ‘rigid’ cataloguing standards (Reference AlstonAlston, 1981a: 373). The implicit tension, then, was between information scientists for whom an isolated record sufficed to stand in for all copies and variations of a handmade book, and scholars for whom this would be a meaningless abstraction rather than a record of a book’s material life history – the purpose of bibliography that W. W. Greg persuasively defined (Reference Greg and FrancisGreg, 1945: 27). As Alston and Janetta put it in their outline of the 18thC STC, ‘the larger purpose, in which that record can reveal contextually its ancestry and its offspring, is for most students of cultural history of far greater significance’ (Reference Alston and Janetta1978: 29).

The sheer scale of the print output of the period meant that the 18thC STC had to delimit its scope in a variety of ways. Alston and Jannetta noted that the 18thC STC aimed to ‘describe a corpus of printing more than ten times the size’ of the two existing STCs of pre-1700 material (Reference Alston and JanettaAlston & Janetta, 1978: 24). It set generic limits to the material and would not include: engraved material; printed forms, such as licences, warrants, certificates, etc.; trade and visiting cards, tickets, invitations, currency (although advertisements were to be included); playbills and concert programmes; playing cards, games, and puzzles (Reference Alston and JanettaAlston & Janetta, 1978: 16–17). More significantly, it constructed its scope along geographic and linguistic lines. It would include:

1. 1. All relevant items printed in the British Isles in any language;

2. 2. All relevant items printed in Colonial America, the United States (1776–1800), and Canada in any language;

3. 3. All relevant items printed in territories governed by Britain during any period of the eighteenth century in any language;

4. 4. All relevant items printed wholly or partly in English, or other British vernaculars, in any part of the world.

Its Anglo-American foundation is clear here, as well as an unconscious legacy of British colonialism in the continuing idea of the British Commonwealth. For Alston the catalogue was also tied to the function of a new British national library. Conscious that the project came on the heels of the British Library’s creation after it was separated from the British Museum in 1973, he felt it was the new institution’s ‘responsibility for the production of the national bibliographic record’ (quoted in Crump, in Reference Snyder, Snyder and SmithSnyder, 2003: 49). Indeed, cataloguing is not a neutral process. Molly Hardy’s reflection on the cataloguing and marginalisation of African-American printers usefully highlights how ‘our organization of data, no matter how neutral we imagine it to be, is built out of and therefore reflects on a particular moment’ and is therefore organised ‘within a system that can never itself be neutral because its creation, like the data it captures, is a humanist endeavor’ (Reference HardyHardy, 2016).

The 18thC STC project was a bold attempt to catalogue in as much detail as possible the ‘ancestry and … offspring’ of thousands of printed works. But it is worth emphasising that a catalogue record is a representation of a book, and the catalogue as a whole was a select representation of the printed output of the period. Moreover, the way in which the electronic bibliographic data of the 18thC STC was able to capture the ‘eccentric evidence’ of hand-press books would directly shape how readers apprehended the nature of the books when they were transformed into microfilm images and, subsequently, into digital images and metadata in ECCO. In addition, the careful delineation of the scope of the 18thC STC would directly shape the content of both the microfilm collection and ECCO.

Books into Images: Microfilming in the Twentieth Century

In 1981 it was announced that the British Library Board had approved the bid from Research Publications (RPI) to ‘reproduce in microform a comprehensive library of eighteenth-century texts based on ESTC’. Robin Alston, the UK editor of the 18thC STC, was, initially at least, the director of the project.Footnote ⁵ Alston envisioned that the 18thC STC online catalogue would be a gateway to the filmed copy: ‘users – whether libraries or scholars – will be given a unique opportunity to acquire access to both bibliographical records and whole-text reproduction’ (Reference AlstonAlston, 1981b: 2). Alston also hoped that microfilm would preserve the fragile, rare books at the British Library from ‘the unequal struggle between book and reader’, anticipating that ‘the proposed microform library will have a tremendous impact on the total conservation effort’ (Reference AlstonAlston, 1981b: 3). Alston’s announcement emphasised access and preservation. His vision for the microfilming of the 18thC STC echoed the technological progressivism of a number of scholars, librarians, and publishers who promoted microfilm as the pre-eminent solution to the preservation of and access to rare books in the twentieth century.

Cultural demands of fidelity and accessibility have driven changes in the technology of reproduction since the 1700s and consequently have raised questions about how old books are represented as visual images (Reference McKitterickMcKitterick, 2013). Key to these issues was the application of photography to creating facsimiles. As David McKitterick argues, by the late 1800s the lure of the image had a profound effect: by ‘seeming to withdraw one of the veils between original and reproduction, replacing human intervention by chemical and mechanical processes, photography offered a new kind of reliability, and even a new kind of truth’ (Reference McKitterickMcKitterick, 2013: 127). The promise of absolute fidelity as a kind of preservation of or even substitute for the original, combined with cheapness and therefore accessibility, would not only spur the production and dissemination of photographic facsimiles of old books in the nineteenth century, but formed the driving arguments of both the twentieth-century microfilming and digitisation projects too.

What led microfilm to be the dominant reproduction technology of the twentieth century? It had been the subject of interest since the invention of photography, although early experiments were perceived as novelties. But it was the confluence of technology and sociocultural forces in 1920s and 1930s America that provided the conditions for the beginnings of the micropublishing industry.Footnote ⁶ One of the most significant moments for this history is the meeting in 1934 between historian Robert C. Binkley, secretary of the Joint Committee on Materials for Research (JCMR), and Eugene B. Power, of Edwards Brothers publishing: it was at this meeting that various developments in microfilming were brought together, and it was arguably the catalyst of a new publishing industry.Footnote ⁷ They discussed the microfilming experiments by R. H. Draeger of the US Navy, and the Recordak Corporation’s ‘Check-O-Graph’ machine that photographed cheques on a continuous 16 mm film, which enabled banks for the first time to verify cheques and to counter fraud (Reference PowerPower, 1990: 23–6). They also discussed the ‘Bibliofilm Service’ established by the librarian Claribel Barnett to copy the records of the hearings of the National Recovery and the Agricultural Adjustment Administrations (Reference BinkleyBinkley, 1936: 134). Binkley described its operation in this manner: ‘a page of print or typescript is photographically reduced twenty-three diameters in size, being copies on a strip of film ½ inch wide and one or two hundred feet long. The micro-copies are rendered legible by projection. A machine throws an enlarged image downward on a table, where the reader finds it just as legible as the original page’ (Reference Binkley and FischBinkley, 1935/Reference Binkley and Fisch1948: 183) (Figure 4).

Figure 4 Microfilming public records, New Jersey, 1937

Hearing about these projects in late 1934 provided Power with his notion of a radical new way of publishing:

Power had already been photographing old books as early as 1931, but systems like Draeger’s or Recordak’s could do this on a bigger scale: the technology of bulk photography, the mass storage of filmed documents, and the ability to print the document – or book – when required would ‘make possible the production of a single, readable book at a low unit cost’ (Reference PowerPower, 1990: 17). It sounds uncannily like digital print-on-demand books of the early twenty-first century available via Amazon. Power began his own microfilm project based on the books catalogued by Pollard and Redgrave’s STC and started operations at the British Museum in 1935. After a number of US libraries took up a subscription service for these films, in 1938 Power left Edwards Brothers and formed University Microfilms Incorporated (UMI), arguably the most successful international microfilm publisher of scholarly materials (Reference PowerPower, 1990: 28–9, 32–5). It is perhaps significant for the history of academic publishing that the large-scale reproduction of old books was undertaken by a commercial business. As we will see in the chapter called ‘Interfacing’, the effect of commercial proprietorship over scholarly materials becomes a significant issue in relation to digitisation projects.

The dream that technology could offer more universal access to knowledge was perhaps most stirringly articulated by H. G. Wells in his 1937 essay ‘The Idea of a Permanent World Encyclopedia’:

However, Reference Binkley and FischBinkley’s 1935 paper ‘New Tools for Men of Letters’ (based on an earlier memorandum of 1934) is even more prescient. It is perhaps the earliest and fullest published reflection on access, scholarship, and new media technology. For both Power and Binkley the challenge was to figure out how to reproduce and distribute low-demand works and difficult-to-access scholarly materials:

While ‘New Tools’ considers a variety of reproduction technologies, microfilm is proposed as the best solution because it ‘offers the reader a book production system more elastic than anything he has had since the fifteenth century; it will respond to the demand for a unique copy, regardless of other market prospects. So the scholar in a small town can have resources of great metropolitan libraries at his disposal’ (Reference Binkley and FischBinkley, 1935/Reference Binkley and Fisch1948: 184).

This last remark about the ‘small town’ scholar is uniquely Binkley’s. His aim in harnessing the new technology of microfilm is to extend access: ‘Let there be included among our objectives’, he argues, ‘not only a bathroom in every home and a car in every garage but a scholar in every schoolhouse and a man of letters in every town. Towards this end technology offers new devices and points the way’ (Reference Binkley and FischBinkley, 1935/Reference Binkley and Fisch1948: 197).

In addition to access, issues of conservation and preservation drove the tremendous expansion of microfilming throughout the twentieth century, such as preserving the rare materials endangered by the wars in Europe (Reference PowerPower, 1990: 117, 125–7), preserving newspapers at the British Museum from the late 1940s (Reference HarrisHarris, 1998: 601–2), and, indeed, conserving the books in the British Library catalogued by the 18thC STC (Reference AlstonAlston, 1981b: 3). In 1988 Patricia M. Battin argued to a US committee that the fragility of books means that ‘it is the record of our shared symbolic code itself that is decaying and endangered. We cannot expect the societal cohesiveness that comes from a symbolic code if the record that comprises it is lost to us’ (Reference BattinBattin, 1988). Technology, it is implied, can save the texts of Western culture. In 2001 there was a burst of public debate about the effects of a mass microfilming project, spurred by Nicholson Baker’s provocative attack on libraries’ disposal of books and newspapers. Baker’s aim was to expose the ostensible fragility and brittleness of books and newspapers as a myth and to reveal the libraries’ microfilming (and digitisation) programmes as a kind of technological zealotry (Reference BakerBaker, 2001).Footnote ⁸

But what was being preserved: the information (the ideas and words) or the material medium (books, documents)? In this light, Robert Binkley’s conception of the scholar’s attitude to the nature of their material is striking. ‘All the documents’ which the scholar uses, he states,

In this conception the medium of reproduction does not affect the content or meaning of the document. The physical nature of the document or book is of little importance: access is everything. At the same time as this rush to microfilm, others were alarmed by what might be lost by the new media technologies of reproduction (Reference GitelmanGitelman, 2014: 63–4). A technology that can reproduce ad infinitum identical images of the original object in the name of accessibility poses questions about the relationship between original and copy. This is the burden of Walter Benjamin’s 1936 essay ‘The Work of Art in the Age of Mechanical Reproduction’. The essay is largely a response to the increasing popularity of film, but it argues that photography has precipitated ‘the most profound change’ on the authenticity of the work of art. This, he says

For scholars of the book the processes of reproduction potentially pose a similar danger to the ‘historical testimony’ of the ‘unique’ object of the book. In 1941 – at the same time UMI was busily filming at the British Museum – bibliographer William A. Jackson had been asked to comment on the technology of microfilm which had been ‘contagiously expounded.’ He enumerated the failures of two-dimensional black-and-white film shot in one plane to capture variations in the tone, colour, or quality of paper, to prevent the introduction of stray marks and blots, and finally, to reproduce the ‘intangible’ features of a book perceptible only by touch (Reference JacksonJackson, 1941: 281, 285). In his 2001 article ‘Not the Real Thing’ Thomas G. Tanselle cited the example of the American preservation project under Battin, pointing out that this project of microfilming ‘was not designed to save books but rather the texts in them’. Tanselle argued that such projects suffer from a ‘confusion surrounding the relation between books (physical objects) and verbal works (texts made of language)’ and suggested a ‘pernicious’ perception ‘that a text can be transferred without loss from one object … to another’ (Reference TanselleTanselle, 2001).

We’ve come to a striking set of parallel worlds: on one hand the scholar-technologists like Power and Binkley, and on the other theorists like Benjamin and scholar-bibliographers like Jackson and Tanselle. Binkley and Power clearly had in mind a method of disseminating costly or hard-to-access scholarly documents. Their aim was not essentially bibliographic – the concerns of historians of the book like Jackson and Tanselle – but ‘informatic’ (Reference GitelmanGitelman, 2014: 63). One question this book pursues is this: how would surrogates such as filmed books be used? As Deegan and Sutherland put it, ‘Under what circumstances or what purposes is a facsimile a satisfactory surrogate for the object itself? … Are we preserving features of the objects themselves or only the information they contain?’ (Reference Deegan and SutherlandDeegan & Sutherland, 2009: 157–8).

The Eighteenth Century: Microfilming the Catalogue

Research Publications was awarded the British Library contract for filming the 18thC STC. Robin Alston had been responsible for assessing bids for the filming in late 1980 from Chadwyck-Healey Ltd., University Microfilms International (UMI), Newspaper Archive Developments Ltd., and RPI (Reference AlstonAlston, 2004: n. 119, 120, 121).Footnote ⁹ Samuel Freedman founded RPI in 1966, in Meckler’s words, ‘for the express purpose of micropublishing significant archives and documents’. Significantly for Alston’s decision, RPI’s series Goldsmiths’-Kress Library of Economic Literature and American Fiction, 1774–1910 included work with library collections and catalogues that comprised eighteenth-century books (Reference MecklerMeckler, 1982: 96, 93).

Alston estimated that the filming of the 18thC STC would ‘take fifteen years to complete’ (Reference AlstonAlston, 1981b: 2). In fact, filming took approximately twenty-eight years between 1982 and 2010 when the programme stopped, and even then, did not manage to film everything in the ongoing and massive cataloguing programme of what was by then called the English Short Title Catalogue (Reference De Mowbrayde Mowbray, 2020a). Alston also envisaged that the bulk of the filming programme would be based on British Library holdings, and so the operation required the secure transportation of hundreds of thousands of eighteenth-century books to and from the British Library in London to RPI’s UK offices in Reading, where filming was monitored by visiting British Library librarians (Reference AlstonAlston, 2004; Reference Bankoski and De MowbrayBankoski & de Mowbray, 2019). The eventual programme would entail filming at many libraries outside the United Kingdom.

The collection was arranged into subject headings and comprised eight categories. These subject headings may well have had their origin in Alston’s experiments with the 18thC STC’s initial online interface at the British Library, which he felt could ‘help in the creation of subject packages which will form the basis of the RPI program to microfilm the substantive texts in ESTC’ (Reference AlstonAlston, 2004). The legacy of this arrangement can be seen in the interfaces of the ECCO and Jisc Historical Texts collections, which can both still be searched by these very same categories:

• Religion and Philosophy

• Literature and Language

• History and Geography

• Fine Arts and Antiquities

• Social Science

• Science, Technology, and Medicine

• Law (Criminal, National, and International)

• General Reference and Miscellaneous

Each reel of 35 mm film was devoted to books from one category. Research Publications sold the collection either in ‘units’, each comprising thirty-five reels, or by subject heading, or the complete collection. In 1994 the complete collection, at that point comprising 178 units, cost £341,760; each unit of thirty-five reels cost £1,920. Individual reels could be bought for £55 or £70 each depending on the total quantity ordered at one time. Given this was a major investment for most libraries, RPI offered ways of ameliorating up-front costs: libraries who took out a standing order of a minimum of one unit per year could obtain a discount and customers could also take out an annual subscription for future units (Research Publications, 1994).

The first films were produced in 1982. While filming rates varied depending on the site, by 1986 it was estimated that books were being filmed at a rate of around 50,000 pages per week at the British Library (Reference De MowbrayDe Mowbray, 2019b). By 1993 RPI had published 6,230 reels and was producing ‘16 units a year’ – that is, 560 reels (Research Publications, 1994). By 2007 the collection was ‘expected to contain over 200,000 items’, and by the time filming stopped in 2010 the collection amounted to 18,094 reels.Footnote ¹⁰

From the beginning the scale of the microfilming programme was a challenge. Given the huge amount of printed material in the period, Alston envisioned that the microfilming of the 18thC STC would be different from UMI’s pre-1700 collection, Early English Books, in which ‘no selection was involved, and every discrete item benefitting from an entry number was, and still is being filmed. This approach could not possibly have been adopted for ESTC: indeed, one pauses to wonder whether libraries (or scholars for that matter) have been significantly helped by the provision of whole-text filming for variants, re-impressions and re-issues, especially when set against the ever-rising costs of production’ (Reference AlstonAlston, 1981b: 2).

Alston proposed that the 18thC STC programme should be ‘selective’, in contrast to the apparent bibliographical promiscuity of UMI’s microfilming (Reference AlstonAlston, 1981b: 2). This policy would change, but at least with the first phase of filming, this vision seems to have been partly carried out. The ‘Research Tools’ section of ECCO’s original interface outlines this part of its microfilm history:

The selection is also an illuminating reflection of who was considered worthy of having all editions of their work microfilmed: in 1982 the scholarly perception of the canon of eighteenth-century ‘major authors’ included no women writers or persons of colour.

In May 1987 RPI announced in a newsletter to subscribers that it would be expanding its filming programme ‘to enrich the potential for research and study offered by the microfilm collection’. The announcement gives a sense of the scale of the new filming programme, expanding enormously beyond Alston’s initial vision: from 1988 ‘the selection criteria will be extended to include all distinct editions of a work insofar as the ESTC bibliographical record makes possible’ (Research Publications, 1987). A 1994 brochure added that the collection also includes ‘variant and pirated editions’ (Research Publications, 1994).

The 1987 announcement is also fascinating since it reveals how RPI identified and responded to a particular shift in scholarship:

Research Publications strategically positioned its decision in relation to the rise of studies in the history of the book, arguably attributable to the influence of the Histoire du Livre of the 1980s led by Roger Chartier and Robert Darnton. Research Publications’ argument rests on the implication that bibliographical records are insufficient on their own: you need access to the ‘actual books’. This is perfectly right, of course; certain aspects of book history and bibliography require the analysis of the physical book. However, RPI’s announcement was artful marketing: while the collection certainly enabled access to a version of a book, the collection could not offer access to a physical book.

Research Publications’ 1994 brochure draws on a similar discourse:

As in Power’s and Binkley’s visions of the 1930s, access to rare research documents was the defining advantage to microfilm. Research Publications’ brochure echoes this affordance, claiming it ‘will bring a comprehensive rare book archive to your library’, as if delivering a research library to your doorstep. Like it had in its newsletter of 1987, RPI trumpets the collection’s ability to aid the scholar interested in the history of the book and in bibliography: ‘Literary scholars can make textual comparisons between variant editions of many works’ and can ‘investigate the mysteries of false imprints or examine the printing history of a particular work’ (Research Publications, 1994). This is true to an extent, but as we will see in ‘Bookishness’, some mysteries need more than a flat, two-dimensional representation of a book to be unlocked.

Access, however, is inextricable from the experience of how a particular technology of remediation is used, and there was a glaring disparity between the commercial praise of the medium and the physical experience of using the technology. It involved fiddling with reels and threading film, then shuttling through the items on each film to find the book you wanted (by hand-crank or motor power); even then, the image could often be marred by poor lighting. I feel the pain of a reader in 1940 who remarked, ‘reading by means of a mechanical contrivance is so new and so unprecedented in the entire history of writing and printing that it introduces, in addition to its real and obvious difficulties, those of a psychological nature on the part of the prospective reader’ (quoted in Reference MecklerMeckler, 1982: 60). As an interface to old books the microfilm reader was not user-friendly.

Figure 5 Microfilm reader and author’s laptop

British Library, 2019

Against the reality of microfilm use, the methods of reproduction promoted by libraries and publishers conjured an image of an easily and immediately navigable landscape of books and information. This is nicely captured in a 1979 study of microform publishing that ends with a vision of the future in which a car would have a microfiche viewer instead of using printed road maps, presumably using one while stationary (Reference Ashby and CampbellAshby & Campbell, 1979: 170–1). The future is here, these studies seem to be saying: witness the title of a study published in 2000 – Micrographics: Technology for the 21st Century (Reference SaffedySaffedy, 2000). Saffedy’s book, to be fair, concludes by considering the impact of digitisation and computers. He proposes a system whereby the use of microfilm collections might be synthesised with ‘electronic document imaging systems’ and ‘database management software’. Perhaps unconsciously echoing Vannevar Bush’s ‘Memex’, Saffedy imagines ‘retrieval stations’ that would comprise a mix of computers for images and database searches and microfilm reader-printers (Reference SaffedySaffedy, 2000: 120). Such an experience was clearly in the mind of Robin Alston when, in his announcement for the microfilming of the 18thC STC, he anticipated that the users could navigate between online access to electronic records and microfilmed images of eighteenth-century books (Reference AlstonAlston, 1981b: 2). His hopes echo the kind of utopianism of Binkley: that technology would enable not only access to rare texts, but also a way of practising scholarship. But how might this practice work when it comes to the study of an actual book from the eighteenth century, navigating between book, catalogue record, and filmed and digitised image?

Book, Record, and Film

In 2002 a camera operator from Primary Source Microfilms arrived at the Kenneth Spencer Library of Kansas University to begin filming books, a process which took place from late May 2002 to early February 2003 (Reference CookCook, 2019). While there, the operator filmed a 1789 copy of Patrick Browne’s The Civil and Natural History of Jamaica. This is the story of that book and how it became a catalogue record and then photographic images, and eventually digitised in ECCO. The story aims to raise some important questions about how we read the relationship between a book’s record – or its bibliographical data, to be more precise – and its digital page images. These questions illuminate the cultural and technological contexts we’ve examined in our prehistory of ECCO.

The Civil and Natural History of Jamaica is famous for being the first book in print by an English speaker to use the classification system created by the pioneering botanist and zoologist Carl Linnaeus (1707–78). Its existence is recorded in the English Short Title Catalogue and represented by two versions of the book: the first edition published in 1756 (ESTC T89757) and a reissue in 1789 (ESTC T89758).Footnote ¹³ Both the 1756 and 1789 issues were filmed as part of the expanded programme from 1988: the presence of both versions in the microfilm collection was probably assured by RPI’s expanded rational to ‘include all distinct editions of a work’ and ‘all variants’ (Research Publications, 1987, 1994). The 1756 issue, according to the ESTC, was not published on film until 2005, and was a copy held in the British Library (shelfmark 459.c.4). However, the first copy of Browne’s History that was filmed was the 1789 copy held in the Spencer Library (call number Linnaeana G13), and it was filmed and published in 2002. The note accompanying the book on the microfilm has a ‘Batch date’ of 10 June 2001, suggesting that the filming of the copy was planned well in advance. Without records of filming operations, it is difficult to say why this copy was filmed before the British Library copy. Possibly of interest was the library’s collection of ‘Linnaeana’ (works associated with Carl Linnaeus) and the fact that the Spencer Library had at that point one of the largest collections of eighteenth-century material in the United States, and so it was probable that this library was made a specific target for filming and that Browne’s History was swept up in the filming of this collection.Footnote ¹⁴

However, the relationship between bibliographical records, this particular book copy, and its filmed and digitised copies illustrates a number of significant issues about how we read what the ESTC catalogue calls ‘Surrogates’. In the ESTC entry for the 1756 edition (T89757) the ‘General note’ states that a ‘variant has pp. 1–12 revised and reset’, yet this clearly relates to the 1789 reissue. When it comes to the physical description in the ESTC catalogue for the 1789 reissue (T89758) there are a number of interesting features. This is the full description:

• Physical description.

• [6], viii, 503, [47]p., plates: map; 2°.Footnote ¹⁵

• General note.

• With a half-title and four additional indexes. With a list of subscribers. A reissue of the sheets of the 1756 edition, with a new titlepage.Footnote ¹⁶

E. C. Nelson argues that, according to his research, the ESTC records are ‘incorrect’ (Reference NelsonNelson, 1997: 333, n. 9). It is true that the description of the 1789 issue does not mention features held in common by all the copies of the 1789 issue, such as the revised and reset pages, that both the 1756 and 1789 issues have a misnumbered page, or the reversed engravings.Footnote ¹⁷ But Nelson misunderstands the nature of a bibliographical description in a catalogue like the ESTC. More particularly, it reveals the challenges the ‘eccentric evidence’ of the unique or individual book copy presents to a universal cataloguing system like MARC (Reference AlstonAlston, 1981a: 379).

To illustrate this eccentricity, there are variations between copies of the 1789 issue itself, which a quick comparison with two other copies available via Google Books reveals. For example, the ESTC states that this issue has a ‘half-title’, but this element is not present in all the copies. In addition, not all copies have the map of Jamaica, and considerable differences appear in the order and presence of the preliminary material.

These variations amongst copies of the same issue of this book bear out Sarah Werner’s remarks about physical descriptions in catalogues, in that they relate to an ideal copy: ‘the imagined version of the book that is most perfect and complete, regardless of whether the library’s copy matches it or not’ (Reference WernerWerner, 2019: 121). In other words, no catalogue could possibly account for all the variations of all book copies held in the world. In this way the ESTC’s description of T89758 is a palimpsest of all the book copies consulted. It has everything: the half-title, the list of subscribers, and the map. To press this point home, not only is the half-title missing from the copy held in the Spencer Library, but the engravings are unusually placed. Rather than being arranged in a numbered sequence towards the end of the book, which is how other copies of T89758 are arranged, in this copy the engravings are scattered throughout the book.

This alone might be enough to make the point about the differences between a record and a unique book copy. However, the book’s appearance on both microfilm and ECCO even more strikingly opens up this gap between a record and a book’s representation in images: nearly every page that has an engraving and an opposite page of text is duplicated. In every instance of duplication there is a faint version of the page and a more defined one. In these page images the first image was also filmed at a skewed angle (Figures 7–10).

Figure 7 ECCO. Image no. 106

Figure 8 ECCO. Image no. 107

Figure 9 ECCO. Image no. 108

Figure 10 ECCO. Image no. 109

Certainly at the point of filming, moving from text to engraved images must have presented a problem in maintaining consistent image quality and contrast. In addition, books of natural history, which often incorporated illustrations and maps, some of which might have been printed on a fold-out page, pose challenges to any camera operator. In this copy our camera operator had three tries at filming the fold-out map of Jamaica (two are partial images, and one manages to fit in the whole map). Less obviously, but perhaps more important in reading the text or if one is relying on ‘all text’ searching, is the fact that six pages are missing.Footnote ¹⁸ In any case it is clear our camera operator was, in the words of one Kansas University librarian, ‘not having a good day’ (Reference CookCook, 2019). The bad day for this one operator was then replicated in the book’s digitisation.Footnote ¹⁹

This finally leads us to consider the refracted relationship between a bibliographical record and an image of a specific book copy. Alston hoped that readers could navigate between these and, to an extent, this was realised in ECCO interfaces where users can pull up bibliographic metadata alongside page images. However, there is an essential tension between the record and the book. Using either the record or the filmed image to read a particular significance into the order of the preliminary material, or into the relationship between the place of the engraved images and the corresponding text, is to confuse the ideal with the material, or the general record with the particular book that was filmed and digitised for ECCO. Michael Gavin has argued that such warnings from scholars are projections of ‘(presumably superseded) naiveté onto the digital project they wished to critique’ (Reference GavinGavin, 2019: 74, n. 5). Gavin, however, presumes a bibliographical knowledge that many humanities students do not possess. Gale never made any claim that individual book images represent the messy, material vagaries of every hand-press book copy in their collections (though EEBO did at one point; see Reference GaddGadd, 2009). However, without some understanding of cataloguing systems, the book in the hand-press era, or the processes involved in the remediation of a book copy, that pitfall exists for the unschooled user.

The aim of this analysis is not to emphasise failings on the part of ECCO or the ESTC or even students and scholars. Instead, this reading of one book from the eighteenth century helps us to understand how human, cultural, and technological factors affect and transform our apprehension of the physical book itself, its bookishness.

Bookishness and the Digital Image

This is a good moment to exemplify how the processes of microfilming interact with the next stage in our books’ lives: the digitisation of the microfilm collection. How we see and conceive these books as digital images in ECCO is to a significant degree dependent on decisions made by RPI/Primary Source Media during the filming programme and by Gale during the digitisation of the film images: these in turn shape our apprehension of these books’ bookishness.

One decision made during the filming of the 18thC STC has profound consequences for understanding these books’ bookishness: no bindings or endleaves were filmed (the blank sheets of paper before and after the text). In the hand-press period these were not part of the printing process, but were the job of the binder, who would usually take direction from the book’s buyer as to the type and quality of the binding. I focus here on one part of the binding, the paste-down (where one endpaper was glued to the underside of the outer binding), because the paste-down is more than merely utilitarian and can be a space for aesthetic decoration (Reference Berger, Duncan and SmythBerger, 2019). Our example is from Charlotte Charke’s autobiography, The Narrative of Mrs Charlotte Charke, and the source copy for ECCO’s first edition of 1755 (ESTC: T68299) held in the British Library (shelfmark G.14246). On the paste-down of the Narrative we can see the owner’s stamp as well as a curious object that seems constructed from scraps of leather bookbinding and looks like a tiny book or the spines of three books on a shelf (Figure 11).Footnote ²⁰

Figure 11 Pastedown, Narrative

The copy belonged to Sir Thomas Grenville; if you look him up you’ll find out that this politician was also a book collector, which might explain the lovely detail of a miniature book. Of course, books were sometimes bound or rebound long after the printing of the books, but nevertheless bindings and endleaves can give us important clues to significant aspects of bookishness, such as its readership or owners (‘provenance’), as well as clues to the context of its production, dissemination, and use (Reference PearsonPearson, 2008: 93–159; Reference Berger, Duncan and SmythBerger, 2019; Reference WernerWerner, 2019: 71–8, 137–8).

Another decision was to photograph books in black and white, a choice clearly based on the processing costs. Colour photography and processing would have increased costs enormously and there was understandably little benefit to be gained, given the fact that mostly all hand-press printing was done in black and white in the eighteenth century. However, it does mean that we miss the chance to see the occasional use of red ink. Compare the two images of this title page from volume one of The Works of Alexander Pope, 1736: one from ECCO, the other from my own copy (Figures 12–13).Footnote ²¹

Figure 12 Works. ECCO

Figure 13 Works. Author’s own copy

Red ink was usually reserved for religious works and almanacs and marked off special days, but red ink was also used sometimes to pick out words on a title page. This simple decision is revealing: title pages were a form of advertisement, so red ink added to the book’s visual impact, heightening the perceived value or status of the book and its author. But given it required the printer to run the sheet of paper through the printing press twice, it was costly in terms of the time and extra patience needed in the printing process: in short, the book better be worth it (Reference WernerWerner, 2019: 57–8). Looking at the title page from my copy of this volume of Pope’s Works, even if we know very little about Pope, the care attending this one page might start to tell us much about either his own or his publisher’s regard for the status and selling power of his works. Granted, the ESTC records the fact that the title page is in red and black and this bibliographic metadata is available in the original ECCO interface and the Gale Primary Sources interface, but we’d have to guess which words since that detail is not recorded. All this most certainly would be lost if we only depended upon the microfilm or the image in ECCO.

It is not just colour that is lost by microfilming. In 1941, in the early years of microfilming old books, William Jackson objected to what I could call the two-dimensionality of film images of books, saying, ‘it is a photograph taken … in one plane’, and adding that the reader can’t benefit from ‘raking’ light (Reference JacksonJackson, 1941: 283). More recently, Sarah Werner made the point that raking light during digital imaging can bring alive the material texture of old books (Reference WernerWerner, 2012). The example I use is again The Narrative of Mrs Charlotte Charke, but this time the second edition of 1755 (ESTC T68298), ECCO’s source copy in the British Library (shelfmark C.184.b.16). In both editions of 1755 the format of the book was duodecimo (12°), a popular, small, and relatively portable format. However, this particular copy is strikingly different: in addition to the printed text of her Narrative, throughout the book are interspersed playbills and newspaper cuttings (Figures 14–15), and there is a lot of blank space around some of the text.

Figure 14 Title page, Narrative, second edition. ECCO

Figure 15 Second page, Narrative, second edition. ECCO

Why is it like this and how was it made? The high-contrast two-dimensionality of the filmed page images in ECCO does not afford the kind of detail that might help. My own photograph of the title page, taken in raking light and from an angle, I hope, reveals more about the book’s construction (Figure 16).

Figure 16 Detail of title page, Narrative, second edition

The edge of ‘original’ printed text is visibly different from the larger-size paper upon which it is mounted. This is a good piece of evidence that we are looking at an example of scrapbooking, where a book owner has collected this miscellaneous material related to Charke’s life and had it interleaved with the printed text of Charke’s autobiography in a book of a much larger format (one of my students called this owner a ‘super fan’ of Charke’s). Double lines appear around the printed text (are these hand-drawn?), and a fainter impression emerges outside these lines (what made this?). Answers to these questions might provide valuable bookish clues as to how it was made, the role of the binder and owner in its construction, and perhaps even how it was used and valued.

Another aspect of bookishness affected by filming and digitisation is the experience of reading an open book. Typically, camera operators for RPI/Primary Source Media would film the books depending upon their size or format.Footnote ²² The smaller book formats like duodecimo and octavo tended to be filmed two pages at a time – that is, they preserved what’s known as the page spread or opening (larger sizes such folios were sometimes filmed one page at a time). As part of the processing of the films into digital image files Gale required single-page images. This processing has a number of consequences for how we see and apprehend books in ECCO.

What we’re looking at now are two consecutive page images from volume six of a 1775 edition, in octavo, of Laurence Sterne’s novel, Tristram Shandy as seen in ECCO (Figures 17–18).

Figure 17 Title page, Tristram Shandy. ECCO

Figure 18 Chapter page, Tristram Shandy. ECCO

The first page image is the title page (as we’ve seen, endleaves and bindings were not filmed). Like in most book digitisations, we are invited to navigate right to see the next page, which is the first page of chapter one. Given that in the West we read from left to right, there is an understandable assumption when we imagine the actual book that the title page is on the left-hand side – the verso – and the next page is on the right – the recto. This is not actually the case, as we can see from a photo of my own copy (Figures 19–20).

Figure 19 Title page opening, Tristram Shandy, author’s copy

Figure 20 Chapter page opening, Tristram Shandy, author’s copy

This might seem a small issue, but it matters for two reasons. First, consider that we experience a book as two pages, or opening, a ‘visual unit formed by facing verso and recto pages in a codex’ (Reference GaleyGaley, 2012). The experience of reading a book in single-page units is a strange distortion of how we actually perceive the printed book as two pages. Second, the blank page opposite a first page of a chapter is not just meaningless, empty space: any printer worth their salt would be expected to give the first page of a new chapter the graphic space it needs. The decisions in digitisation have altered our apprehension of this book’s bookishness.

The examples of what’s happened to these books, and perhaps even the books I’ve chosen, might seem mundane or even unexceptional. But it is precisely because these examples are not exceptional that they are entirely representative, and what can happen can be so easily overlooked. Moreover, I’ve highlighted some of the ‘losses’ to a book’s bookishness during these remediations not as some kind of total critique of ECCO, but in order for us to understand the decisions and processes in ECCO’s history and so help us to understand what we are seeing via ECCO and how we might better use this digital archive.

Publishing, Technology, and ECCO, 1990–2004

It wasn’t called ECCO at first. The original name during the project’s development in 2002 was ‘The Eighteenth Century – Complete Digital Edition’, closely associating it with the microfilm collection on which it was based (Reference QuintQuint, 2002). The finished product, renamed ‘Eighteenth Century Collections Online’, was published by Thomson Gale in 2003 and completed in 2004. It was officially launched in the United Kingdom at the British Library on 24 July followed by a reception at Dr Johnson’s House, London (Reference De Mowbrayde Mowbray, 2019c). In the United States it premiered in early August at the thirty-fourth meeting of the American Society for Eighteenth-Century Studies (ASECS) in Los Angeles.

But let’s go back a little. Thomson Gale’s entry into digital publishing is a good example of the forces at work in the 1980s and 1990s that influenced the shape of commercial academic publishing. First, techno-commercial developments took place in digitisation and in electronic transmission, enabled by cheaper computing power and the rise of personal computing. Second, library budgets faced significant pressures as a result of the increasing prices of journals and the falling demand for monographs. Taken together, these technological and market contexts pushed academic publishers to diversify their product portfolios by merging companies with expertise in electronic publishing or with existing content. These commercial strategies were also a response to the explosion and spectacular failure of companies experimenting with digital platforms and electronic publishing in the 1990s – later dubbed the ‘dot.com bubble’ – arguably influenced by the radical business theory propounded in Bower and Christenson’s essay ‘Disruptive Technologies: Catching the Wave’ (Reference Bower and ChristensenBower & Christensen, 1995; Reference ThompsonThompson, 2005: 81–110, 309–12).

The story of Thomson Gale itself starts in 1998, as the result of a merger involving the long-established company of Gale Research (founded in 1954 by Frederick Gale Ruffner and sold to Thomson in 1985), Information Access Co., and Primary Source Media: this became the Gale Group, whose new headquarters was to be in Farmington Hills (Reference McCrackenMcCracken, 1998). It was a significant moment that combined three companies with particular but overlapping expertise: Gale’s specialism was reference books in the humanities, sciences, and technology; Information Access Co. specialised in CD-ROMs, microfilm, and online information; Primary Source Media was formerly Research Publications (RPI), who, as we know, produced microfilm collections of rare research materials including The Eighteenth Century. As one contemporary observed, the merger was ‘a move towards the electronic age and it helps to be big in this industry’ (quoted in Reference McCrackenMcCracken, 1998).

In contrast to the experimentation of digital companies in the mid-1990s was the more cautious strategy publishers adopted of diversifying and monetising their existing portfolio of resources by transforming them into a new medium. Former Gale CEO Dedria Bryfonski noted that their existing content formed ‘the backbone of electronic products widely used today’ (quoted in Reference EnisEnis, 2014). Thomson had decided to begin digitising its collections as early as 1996, just before the 1998 merger: its first was The Times Literary Supplement Centenary Archive, 1902–1990, published in 1999. ‘Almost all our growth now is from the migration from print to digital,’ as Thomson Gale’s CEO Gordon Macomber pointed out in 2004 (quoted in Reference BerryBerry, 2004). In another 2004 interview he emphasised that the ‘lion’s share of our business is nonaggregated – that is, the distribution of our proprietary content online’ (quoted in Reference HaneHane, 2004). Distributing its specialised (‘nonagreggated’) content enabled Gale to monetise its existing (‘proprietary’) investments, of which ECCO was the prime example.

Thomson Gale was of course eager to emphasise the affordances of digitisation for its products. Singling out ECCO as the exemplar of Thomson Gale’s digital products, Macomber stressed digital accessibility over print, noting that scholars ‘would have had to go [to the British Library] to get access to this material’ (quoted in Reference BerryBerry, 2004). But ECCO was promoted as more than just access; it was represented as offering a new way of learning: as Mary Mercante, vice president for marketing put it, ECCO was also about ‘searchability’ (quoted in Reference SmithSmith, 2003). Creating new markets by diversifying and transforming content also meant harnessing the unique affordances of digital technology. So what technologies of digital reproduction and publishing were available when Thomson Gale was considering digitising its collections in the late 1990s? What was the significance of these choices, and what were Thomson Gale’s competitors doing with their digitisation programmes? And what does ‘digitisation’ even mean?

In its broadest definition digitisation is the conversion of analogue information into digital form, such as a continuously varying voltage into a series of discrete bits of information.Footnote ²³ The forms of digitisation that interest us here are methods that convert text and those that convert photographic images. Text is converted to digital form by encoding each character as a binary string of zeroes and ones, a series of discrete bits of information; for example, the letter ‘a’ would look like this: 01000001.Footnote ²⁴ Photography, as we’ve seen, has been a key media technology for the representation of old books, but while photographs represent continuous gradations and variations in tone, shade, and colour, the digital image comprises discrete bits of information that approximate the effect of continuity. Images are transformed into a grid of discrete pixels, or bitmap, with each pixel assigned a number for colour and intensity. Black-and-white, or bitonal, images are the simplest binary forms, but 8-bit or 24-bit images can encode huge numbers of colours.

Thomson Gale began thinking about digitising its existing collections in 1996, and anyone contemplating digitising old books in the 1990s had a number of technological and commercial decisions to make. They would have to decide whether to transform the material into page images, or to produce it as searchable text, or to offer both: this would depend on the nature of the content, on who their users were, how they were expected to use it, and costs. If searchable text was on the cards, then they would need to decide between two options: to either have each word transcribed by hand (accurate but costly), or to use text automatically generated from page images using software (cheaper, but inaccurate, especially if the content was microfilm). In parallel, there were also decisions to make about how to publish the material: was it to be on CD-ROMs, or would it be online via the nascent World Wide Web?

Humanities computing had a history going back to Roberto Busa’s work in the late 1940s, and digitisation projects had been running since Project Gutenberg’s first electronic transcription in 1971 (Reference Hockey, Schreibman, Siemans and UnsworthHockey, 2004; Reference Deegan and SutherlandDeegan & Sutherland, 2009: 119–54; Reference JohnstonJohnston, 2012; Reference ThylstrupThylstrup, 2018: 1–31). However, Thomson Gale’s attention in the 1990s was fixed on its immediate competitors in the commercial market for educational and academic resources; its engagement with the scholarly field that became known as ‘Digital Humanities’ did not become significant until the 2010s. Google Books, of course, was perhaps the most high-profile mass digitisation project, and was launched at the Frankfurt Book Fair in 2004. The extent to which Thomson Gale knew about Google Books while developing ECCO is not clear, but it’s revealing that in 2004 CEO Macomber chose to characterise Google as an information search-and-retrieval business. Commenting on the ‘challenge … laid before us by Google and Yahoo! and others that dominate the Internet space’, he argued that:

Gale explicitly positioned itself by describing what it offers as ‘differentiated content’ – such as digitised book collections – and a ‘tool to help users learn’ in opposition to the mere aggregation of information by platforms like Yahoo! and Google.

The more immediate models for ECCO were digital products developed by Thomson Gale’s peers and competitors in academic publishing, so the rest of this section looks at two examples: ProQuest’s Early English Books Online (EEBO) and Chadwyck-Healey’s English Poetry Full-Text Database; both are from the 1990s but model different technological and publishing choices. In the 1980s and 1990s, in addition to microfilm and the Internet, there was another publishing technology: the CD-ROM. An adaptation of the compact audio disc, the CD-ROM (compact disc – read only memory) was designed in 1982 as a storage medium for data. Many of the microform and information publishers adopted the CD-ROM format as a way of distributing electronic resources in the 1980s and 1990s, including Gale and Primary Source Media, and even after the advent of interconnected networks of computers the CD-ROM was considered superior to the slow download speeds and expensive connection costs of the early Internet. Its high memory capacity and multimedia capabilities made it an attractive medium for a variety of resources. For example, the 18thC STC project felt that since it ‘had pioneered the automation of rare books cataloguing … compact disc (CD-ROM) was therefore an attractive and appropriate medium’ (Factotum, no. 30, 1989: 3).Footnote ²⁵

The CD-ROM, therefore, was an interesting transitional publishing medium. A case in point was Chadwyck-Healey’s pioneering English Poetry Full-Text Database, published 1992–4. As Chadwyck-Healey relates, the sheer scale of the project was unprecedented: 165,000 poems. ‘We were to turn all of English poetry into one huge, fully searchable database’ (Reference Chadwyck-HealeyChadwyck-Healey, 2020: 231, 217). To enable the user to search for authors, titles, or first lines, every single poem had to be transcribed and encoded by hand in order to accurately render particular elements of each poem. Eventually, Chadwyck-Healey published nine different literary databases this way (Reference ThompsonThompson, 2005: 395). However, the English Poetry package illuminates some of the oddities of CD-ROM publications. On one hand, the ability to conduct searches across the landscape of English poetry demonstrates a crucial affordance of electronically transcribed text. However, the package of disks, software disks, a manual, and a printed bibliography created limitations on its user friendliness. Moreover, the CD-ROM could not meet libraries’ demand to enable multiple users to simultaneously access resources (it could only be used by one person at a time), and it was hampered by both proprietary software and hardware requirements that were often out of date within a few years.

Chadwyck-Healey ruefully acknowledged that ‘in less than ten years, the CD-ROM revolution was over’ (Reference Chadwyck-HealeyChadwyck-Healey, 2020: 260). The CD-ROM represented a form of technology in a state of short-lived infancy: online, networked, and web-based publications were fast becoming the future of academic digitisation products. Networked, multiple-user access, and the ability to update and tailor specific packages for libraries were attractive advantages. Chadwyck-Healey, for example, had to create business cases based on a model of sharing information that was, on the surface, free via the Internet. The answer was an annual subscription charge for access to a resource, as opposed to a library owning it, and Chadwyck-Healey launched Literature Online (LION) in 1996 (Reference ThompsonThompson, 2005: 394–9; Reference Chadwyck-HealeyChadwyck-Healey, 2020: 261–3). This business model was one that companies like ProQuest and Gale quickly adopted.

ProQuest’s Early English Books Online (EEBO) was published as an online resource in 1998 and now comprises 146,000 titles from the period c.1475–1700. In terms of its source material – books from the hand-press era that had been microfilmed – it is similar to ECCO. Early English Books Online was the digitisation of UMI’s original microfilm collection, Early English Books (which began with Eugene Power’s filming operation in the 1930s), and was largely based on the two seminal Short Title Catalogues of printed material from c.1475 to 1700 created in the early twentieth century.Footnote ²⁶ ProQuest had begun to digitise the films in 1998 as black-and-white or bitonal images, and later began processing them in greyscale in order to maximise fidelity to the printed original, although as Mak has pointed out, and as we have seen in our bookish vignettes, the emphasis on digital processing potentially ‘elides the key intermediary of microfilm’ (Reference MakMak, 2014: 1519). Equally interesting as its claim of fidelity was ProQuest’s decision to offer just page images: there was no attempt to use OCR software to generate text from the images. In 1994 UMI were developing its own OCR software, but felt that good-quality OCR of microfilm was twenty to twenty-four months away and still too expensive (Reference SchonfeldSchonfeld, 2003: 42–3). Indeed, the massive JSTOR project to digitise academic journals, developed in 1994, reveals parallel concerns. The CD-ROM limited access, and so JSTOR opted for online full-text searchability. Hand transcriptions were costly, but given the limitations of OCR software at that moment, the project opted to scan the paper originals (Reference SchonfeldSchonfeld, 2003: 11–14, 28–30, 72–3). In response to the perception that EEBO ‘was not quite utilizing all of the potential that electronic technology had to offer’, the University of Michigan Library stepped in to form the Text Creation Partnership (TCP), a remarkable collaborative project that included ProQuest, the University of Oxford Bodleian libraries, and the Council on Library and Information Resources (Reference MartinMartin, 2007: 161). The project aimed to produce text that was accurate, accessible, and reusable: in the chapter ‘Interfacing’, we see how the TCP worked with ECCO, but its initial effect was to enable the kind of full-text searchability EEBO’s original digitisation lacked.Footnote ²⁷

It is significant that Thomson Gale chose to make the technology of ECCO’s searchability a key selling point – and also its scale – since it viewed ProQuest’s EEBO as its key competitor, despite the fact that it offered material from a different period (Bankoski, 2019c). Clearly, by 2000 when the ECCO project began, Thomson Gale felt that the cost-benefits of using OCR software on microfilm were in its favour. As the writer of a document entitled ‘ECCO Prototype Walk-Through’ noted, full-text searching was ‘a MAJOR feature as it searches the full 29 million pages in ECCO; also not available in EEBO’ (unpublished, c.2003; emphasis in the original). Gale was certainly keen to talk up the potential of combining searchability with scale. The 2004 User’s Guide described it this way:

ProQuest’s EEBO comprised 125,000 titles at its launch in 1998: Thomson Gale’s emphasis on the size of ECCO was certainly with its competitors in mind.Footnote ²⁸

Scale and technology cost money: one scholar described it as ‘this quarter-million-dollar investment’ (Reference GarrettGarrett, 2007: 71). Given this, and the fact that ECCO was to be an online product, Thomson Gale’s sales strategy was to offer ECCO as both a product and a subscription. There was a one-off payment for the content and a yearly hosting or access fee: libraries who bought the content could theoretically host the content on their own servers and interface, but no library has yet gone down that route. Reviewers at the time quoted a list price of $500,000 or £315,000 for the complete collection, although ECCO could be bought on a unit-by-unit basis, following the same categories that grouped the microfilm collection, which meant that institutions could buy individual units or spread the cost over time (Reference LevackLevack, 2003; Reference SchwarzSchwarz, 2004). In addition, Gale offered discounted prices for institutions that had bought its microfilm collection (Reference SmithSmith, 2003).

Thomson Gale, therefore, was navigating between costs, existing technological developments, the nature of the original material, the needs of its market, and its competitors. ‘We own the 18th century’; so proclaimed Mark Holland, a UK representative from Thomson Gale (quoted in Reference QuintQuint, 2002). In one sense, this comment reflected Thomson Gale’s literal ownership of the film collection, but in its confident aggrandisement it spoke volumes about Thomson Gale’s ambition to be the one publisher everyone goes to for eighteenth-century research materials.

From Film Image to Metadata: Digitising the Eighteenth Century

Against this background, then, Thomson Gale aimed to create a new market for its content by combining the long-standing bibliographic sentiment of access to old books with the informatic sensibility shaping the online world. The following section explores the three aspects of the books in ECCO: the digital page image, the OCR text, and finally the metadata upon which the interface’s search functions depend.

The digitisation of the microfilm collection began in 2000 (Reference BankoskiBankoski, 2019a).Footnote ²⁹ This was while the programme of microfilming eighteenth-century books was still going, and while the film collection was still being published (filming stopped in 2010). This explains why Gale published a second collection called ECCO II in 2009 of an additional 46,000 or so extra titles: in effect, a scanning programme that caught up with some of the microfilming done between 2003 and 2009.Footnote ³⁰ Gale contracted the digitisation of the microfilm to HTC Global, an IT services company founded in 1990, and whose headquarters is in Troy, Michigan. HTC has offices across the globe, but the microfilms of The Eighteenth Century were shipped out to India for processing. Natraj Kumar, in charge of the operation in India, estimates that nearly 500 people were involved in the two-year digitisation, including more than 350 production operatives, more than 100 on quality control, and 75 in supporting service roles (Reference KumarKumar, 2019). According to Ray Bankoski, the vice president of content and metadata at Gale, most of these workers ‘were keying and QC operators that captured the metadata from the page images. There was an initial ramp up period but we processed about 500 K [page images] per week over the majority of the two year period’ (Reference BankoskiBankoski, 2019a).

The actual conversion of the reels of film images to digital files was accomplished by a Sunrise Model 2000+ scanner and then processed in various ways (Reference KumarKumar, 2019). Typically, microfilming meant photographing a double-page spread or ‘opening’, except in the case of large-format books or books with big or fold-out maps and illustrations where single pages were photographed individually. However, when these were digitised the images were processed into single pages: this was a ‘requirement from Gale at that time. Gale wanted “Clean cropped Page Images” to be delivered’ (Reference KumarKumar, 2019). In addition, films were digitised as bitonal images. These decisions likely helped to produce page images with fewer extraneous marks and a high contrast between the black text and the white page, and therefore aided the OCR software to decode the text. As we’ve seen, the effect on how printed material looks is significant.

Page images were saved as tagged image file format (TIFF) files. This file format was developed in the late 1980s as a standard for storing raster graphics or bitmaps so that digitised images can be easily converted into pixels on the computer screen. For users who want to download page images to their own computers the files are made available as portable document files (PDFs). The PDF was designed in the early 1990s in order to create ‘a universal way to communicate documents’ and ‘should be viewable on any display and should be printable on any modern printers’ (Reference WarnockWarnock, 1991). Developed as part of the desktop publishing revolution that paralleled the arrival of the personal computer, this format has a wider significance in that it is a continuation of the longer history of reproducing books as images: they ‘look as if they work like print’, as Lisa Gitelman argues. Its ‘portability’ aimed to ensure that the image looks the same whether on screen or in print. In this sense, the PDF ‘appeals to the fixity of print’ (Reference GitelmanGitelman, 2014: 114, 118). Such an imaginary materiality would have been a natural fit for databases of old books like ECCO. Project Camelot had even imagined when ‘[l]arge centrally maintained databases of documents could be accessed remotely and selectively printed remotely’ (Reference WarnockWarnock, 1991). This was one of those instances when the social changes wrought by personal computing and the commercial and technological standards developed to meet those demands in the 1990s are part of the reason why old books look the way they do on databases like ECCO. However, the relatively large size of image files (compared to text files) and the limited bandwidth of the early Internet posed certain limitations on the user experience.Footnote ³¹ The concern was, as Julia de Mowbray put it, ‘that a higher page count would adversely hit the performance and speed of the product’ (de Mowbray, 2020a). At its launch in 2003 users were allowed to download only 10 pages at a time; by 2008 this had increased to 50. In 2009 the download capacity was increased to batches of 250 pages at a time (Reference Golderman and ConnollyGolderman & Connolly, 2008: 25; Reference De Mowbrayde Mowbray, 2020a).Footnote ³²

Gale had more in mind for its users than merely downloading and printing off a few pages of, say, the first edition of Robinson Crusoe as a PDF file. Gale wanted to bring the kind of functionality enabled by converting images of its books into searchable text using OCR software. But how does OCR work? Image file formats are opaque to the computer: they cannot be searched for text since they are designed to enable a computer to render that encoded image on a screen. Optical character recognition software effectively scans the image files to ‘read’ the text of each page and convert it into a text file. Optical character recognition technology was first developed in the 1950s, although until the development of cheap powerful computing in the 1990s, it was quite limited in its abilities (Reference Tanner, Muñoz and RosTanner et al., 2009). The OCR conversion process involves several steps. The OCR engine first preprocesses images by making sure the page is aligned correctly and the lines of words are horizontal. It then divides the page into zones, identifying blocks of text such as paragraphs, lines, and words. It then divides these blocks up again to find individual characters. It is perhaps for this reason that the creators of digitisation projects like Gale’s ECCO chose to very closely crop microfilm images into single pages, avoiding the possibility that the OCR engine might mistake stray marks or a page edge for a text block or a character. ‘Once the characters have been singled out’, as Rose Holley describes it, ‘the program compares them with a set of pattern images stored in its database. It analyzes the stroke edge, the line of discontinuity between the text characters, and the background. Allowing for irregularities of printed ink on paper, each algorithm averages the light and dark along the side of a stroke, and advances numerous hypotheses about what this character is. Finally, the software makes a best guess decision on the character’ (Reference HolleyHolley, 2009).

The OCR engine then compares all the characters in a word block to its dictionary of complete words. The text for ECCO phase I material was produced using an OCR engine called PrimeRecognition, which used several different OCR engines operating at the same time with a voting system choosing the best character recognition. For the texts of ECCO phase II, published in 2009, Gale upgraded the OCR engine to ABBYY (Reference BankoskiBankoski, 2020).Footnote ³³

Gale’s original interface for ECCO, like Google Books, showed only page images, and not the completed text output from the OCR engine. If the user searches for a word, how then does this lead them to see the results on a page image? This necessitates a further step: each word is given a specific coordinate so that it matches the location of the word on the page image. These coordinates and the OCR text output are stored in XML format. The example that follows is the first line from Daniel Defoe’s An Essay on the Regulation of the Press (1704):

<wd pos="226,548,290,588">LL</wd>
 <wd pos="308,547,393,587">Men</wd>
 <wd pos="414,540,561,596">pretend</wd>
 <wd pos="589,540,651,586">the</wd>
 <wd pos="675,537,949,586">Licentiousness</wd>
 <wd pos="962,537,1016,585">.of</wd>
 <wd pos="1033,538,1097,583">the</wd>
 <wd pos="1072,523,1113,596">:,</wd>
 <wd pos="250,599,346,641">Press</wd>
 <wd pos="363,605,402,640">to</wd>
 <wd pos="423,597,469,638">be</wd>
 <wd pos="495,606,516,639">a</wd>
 <wd pos="544,593,689,647">publick</wd>
 <wd pos="702,591,914,642">Grievance,.</wd>

<page type="bodyPage" firstPage="yes">
 <pageInfo>
 <pageID>00030</pageID> <assetID>3304394388</assetID>
 <ocrLanguage>English</ocrLanguage>
 <sourcePage>3</sourcePage>
 <ocr>89.1</ocr> <imageLink pageIndicator="single" width="1184" height="2080" type="tiff" colorimage="bitonal">014720280000030.TIF</imageLink>
 <!--image src="065600_2.tif" -->
 </pageInfo>

If you look back to the OCR text and the word coordinates for the first line from page three you’ll see that the line ‘All Men pretend the Licentiousness of the Press to be a publick Grievance’ contains some errors: the OCR software has introduced stray punctuation marks and rendered ‘All’ as ‘LL’. Despite the developing technology of OCR, it’s only as useful as its accuracy in word recognition. In the case of ECCO, the accuracy and reliability of a user’s search results depend entirely on the accuracy of the OCR engine to decipher eighteenth-century print.

More sophisticated text recognition can be achieved by an OCR engine that can match alternative spellings or find similar forms of a word (known as ‘fuzzy’ searching) – particularly important for printed texts published before the nineteenth century, which used a wide variety of spellings and word forms. The period of hand-press printing also produced idiosyncratic typography. For example, printers used the so-called long s (which superficially looks like an ‘f’), ligatures (when two characters, for example, ‘f’ and ‘l’, are joined together as a single glyph on one single piece of type: fl), and the swash, or exaggerated serif. Also problematic are pages that combine text with images, tables, or diagrams. Such peculiarities are exacerbated by the huge variations in the quality of the material itself and the sheer human fallibility of handmade printing processes (such as skewed lines, ink bleed-through, under-inking leading to faded text, using worn or damaged metal type). All this means that OCR engines had and still have a real difficulty discerning the features of pre-1900 print.

If that were not enough, the fact that the OCR engine is reading a book that is twice remediated – first by filming, and second by digitisation – introduces more barriers to the software’s ability to render accurate text. Microfilming itself was not always an error-free process: witness JSTOR’s decision to scan from the original paper documents and not from microfilms. Although quality control procedures were standard, a number of faults could be introduced at the filming stage, some of which could be rectified, and others not. These included spots or areas missing from faulty film, blurred images, missing pages, foreign objects showing, such as the operator’s finger, inconsistent exposure resulting from changes in the density of the pages, and scratches on the film from grit in the camera (Reference Ashby and CampbellAshby & Campbell, 1979: 81). A 1990 report by the University of California, Berkeley evaluated five US microform vendors, including RPI, and concluded that while most microfilming companies had basic acceptable standards, all had some minor faults, and RPI had a ‘major fault’ of missing pages (Reference Lockhart and SwartzellLockhart & Swartzell, 1990: 120–3). Indeed, any full-text search of the 1789 issue of Browne’s The Civil and Natural History of Jamaica will not find any words on the pages missing from its filming.

It is not surprising, then, that OCR is the most well-known topic in discussions of ECCO (Reference GaddGadd, 2009; Reference GreenfieldGreenfield, 2010; Reference SpeddingSpedding, 2011; Reference DaneDane, 2012; Reference MandellMandell, 2015; Reference HineHine, 2016; Reference Prescott, Brandtzæg, Goring and WatsonPrescott, 2018; Reference Hill and HengchenHill & Hengchen, 2019). For example, Ian Gadd revealed that ‘the word “fuck” or versions thereof appear over 28,000 times,’ revealing the OCR engine’s tendency to misread the long ‘s’ as an ‘f’. It is unlikely that eighteenth-century printed material was much cruder in its language than we thought; in other words, the OCR engine has produced an alarming number of false positives (Reference GaddGadd, 2009:10).Footnote ³⁶ Moreover, while OCR software begins at the level of the character, a 2009 report on the OCR accuracy of the British Library’s digitisation of newspapers remarked that the ‘potential retrieval rate for a resource depends upon the OCR engine’s accuracy with significant words, that is, those content words for which most users might be interested in searching, not the very common function words such as “the”, “he”, “it”, etc.’ (my emphasis).Footnote ³⁷ The report found that the OCR accuracy with significant words was always lower than character or word accuracy rates, which ‘may be due to such words being generally longer or not in dictionaries, which thus provides more statistical opportunity for error’ (Reference Tanner, Muñoz and RosTanner et al., 2009).Footnote ³⁸ Optical character recognition accuracy can be calculated in different ways; later projects examined accuracy at the level of the page. For example, the Early Modern Optical Character Recognition Project (eMOP), which explored ways of improving OCR in both EEBO and ECCO, calculated ECCO’s OCR ‘correctness’ at the level of the page as 86 per cent (Reference MandellMandell, 2015: 2). However, the most recent and most systematic analysis revealed that its accuracy at the character level was very good, but at the page level it was noticeably lower at 77 per cent.Footnote ³⁹ The analysis emphasised that ‘OCR errors are not neutral’ in ECCO since the peculiarities of eighteenth-century typography were a major factor in error and disproportionally affected long words, so perhaps the more significant words (Reference Hill and HengchenHill & Hengchen, 2019: 828, 840, 829–31).

To an extent, the precise figure for recognition accuracy is a distraction; the real issue is one of transparency. To be fair, Gale has always acknowledged that ECCO’s OCR is fallible. On the original interface it was rather hidden in ‘FAQs’:

The note perhaps managed to both understate the problem and overstate the solution. Fuzzy searching might help to find additional matches, but it can’t entirely overcome the challenges posed by the material and the processes by which it was initially filmed. Gale’s initial decision to not make the underlying OCR text visible to users on the original interface was perhaps due to the fact that the OCR text ‘contains enough errors to be distracting to the reader’, as the TCP diplomatically put it.Footnote ⁴¹ In the chapter ‘Interfacing’, we’ll see how Gale changed its stance towards the OCR text, but in this initial period the fact that the actual OCR text was not accessible in the original ECCO interface raised certain problems, particularly the necessary knowledge needed on the part of users. Increasingly scholars were using the large sets of textual data to conduct computer-aided quantitative analyses for posing questions about themes or topics, authorial style, or linguistic features, all of which depend, in essence, on counting words. But the accuracy of such counting crucially depends on knowing the quality of the textual data in order to know the limits and contexts of any results. A case in point was Peter de Bolla’s fascinating study The Architecture of Concepts: The Historical Formation of Human Rights (2013). This used ECCO’s interface to find keywords and the co-occurrence of other keywords to show how the concept of ‘rights’ shifted over the period. His note on methodology is interesting:

Some responses to this study have pointed out de Bolla’s overconfident reliance on ECCO’s online search capabilities, and the inconsistencies in his note on methodology that acknowledges the fact that the OCR ‘compromises the reliability of the data’, but at the same time is confident that this will not compromise the conclusions (Reference BakerBaker, 2016; Reference Prescott, Brandtzæg, Goring and WatsonPrescott, 2018: 63–4). The claim to the usefulness of the book’s conclusions may be true: even with 80 per cent accuracy, quantitative analyses can still produce meaningful results (Reference Hill and HengchenHill & Hengchen, 2019: 840). However, without direct access to the textual data, and without taking into account the exact nature of your eighteenth-century data set, how does the researcher know what wasn’t found. In other words, you don’t know what you don’t know.

Searching the OCR text of the books in ECCO isn’t the only way – and perhaps not even the first way – users navigate the database. As I asked in 2007, ‘Could students begin to use it as a research tool for themselves?’ (Reference GreggGregg, 2007). Going by the second wave of responses to ECCO around 2007–9, in conference panel discussions and articles, academics and librarians assessed how they and their students reflected on the relationship between access, discoverability, and searchability. Eighteenth Century Collections Online arguably supported the turn towards historicisation which emphasised the importance of contextual material in understanding literary works; a related consequence was that courses could introduce texts that pushed the boundaries of the canon, illuminating, for example, neglected genres or writings by women (Reference AwAw, 2007; Reference GreggGregg, 2007). A special edition of The Eighteenth-Century Intelligencer, ‘Teaching with ECCO’, explored how students could access texts held in libraries’ special collections and that are often inaccessible either financially or geographically; it also discussed how ECCO posed some initial challenges to students unfamiliar with database searching and the nature of such a large collection of historical material.

Given what Sayre Greenfield called this ‘huge wilderness’ of digitised primary material, how did ECCO’s various search functions shape the way users navigated this wilderness or enable discovery (Reference GreenfieldGreenfield, 2009: 13)? I want now to turn to how ECCO’s original interface worked beneath its surface (Figure 21).

Figure 21 Original, stand-alone interface 2009–2021

(screenshot, 31 May 2020)

As Lori Emerson reminds us, ‘interfaces themselves and therefore their constraints are becoming ever more difficult to perceive because of the blinding seduction of the wondrous’: the interface itself is not a neutral or transparent window to the books in ECCO (Reference EmersonEmerson, 2014: ix). Eighteenth Century Collections Online is not just what you see on the screen; it is a collection of files and systems that drive the capabilities of the interface: it shapes what the user is capable of doing with the books (which are themselves made up of image files, OCR text, and their associated bibliographical data). The search functions on ECCO’s original interface were driven by a search engine based on Apache’s Lucene. This might seem an arcane detail, but knowing that search software is an essential part of ECCO’s workings also means that we might want to ask: when we search ECCO, what exactly is happening?

A 2003 prototype walk-through explained its search capabilities. From the ‘Basic Search’ box (later called ‘Find’, emphasising the aspect of discoverability) there were options to search four fields: the full text of the books, author, title, and keyword. We have already discussed the full-text search of the OCR’d text. Searches for author or title are enabled by the way in which the metadata for each book – essentially from the ESTC record – is structured in its XML file. In other words, the author or title search field searches for the author or title ‘tag’, as we can see from our example XML:

    <authorGroup role="author">
    <author>
<composed>Defoe, Daniel</composed>
<first>Daniel</first>
<last>Defoe</last>
<birthDate>1661?</birthDate>
<deathDate>1731</deathDate>
      </author>
    </authorGroup>
    <titleGroup>
<fullTitle>An essay on the regulation of the press</fullTitle>
<displayTitle>An essay on the regulation of the press</displayTitle>
    </titleGroup>

Users could, in addition, limit their search by defining ‘Subject area’. This was always a confusing category, since it implied a notion of a book’s ‘subject’, but this actually reflected the way in which the original microfilm collection was categorised and sold, and the subsequent publication of ECCO itself (such as ‘Literature and Language’, or ‘Fine Arts’, or ‘History and Geography’). The process of adding metadata in ECCO also included tagging, which defined the following illustration types:

• Cartoon

• Chart

• Coat of Arms

• Genealogical Table

• Illustration

• Map

• Music

• Plan

• Portrait

This in turn enabled the facility in ‘Advanced search’ to limit searches by specific types of illustration.

However, how do we know what we’re looking for before we discover it? The ability to discover new and unfamiliar texts using ECCO’s ‘keyword’ search function was felt to be one of its most powerful capabilities to enable a kind of serendipitous navigation of this huge database (Reference GreenfieldGreenfield, 2009). In this option the search engine’s algorithm was actually searching across multiple fields:

• Author (including any secondary author fields)

• Title (and all variations of the title – short title, full title, variant title, etc.)

• Chapter headings, which is essentially equivalent to the info that you see in the e-TOC (minus the three major headings: Front Matter, Main Body, Back Matter)

• Library of Congress subject headings (when available in mid-2008)Footnote ⁴²

The chapter and section headings were scanned as part of the digitisation and were used to create an ‘e-Table of Contents’ (e-TOC) in order to enable navigation within each text, but were also tagged in the XML file so they would be searchable. The last field mentioned in the keyword search, ‘Library of Congress subject headings’, refers to the original interface’s search capabilities from 2008 onwards. From this date the interface offered one more additional field to its Basic Search or Find box: ‘Subject’.

This wasn’t the same as Gale’s subject areas, but it was a library subject heading, and it vastly enhanced the usefulness of ECCO’s keyword searching. The Library of Congress maintains the classification system for subject headings that is the standard for library cataloguing around the world.Footnote ⁴³ Subject headings are a form of bibliographic metadata that attempts to categorise a book’s intellectual content, rather than details about its physical form. For example, the subject headings for Patrick Browne, The Civil and Natural History of Jamaica (1789) are:

• Tropical medicine – Early works to 1800

• Medical climatology – Early works to 1800

• Natural history – Jamaica – Early works to 1800

Subject headings offer the user the ability to discover material based on what might initially be a mere hunch, without any necessary knowledge as to titles or authors of the material. It was this added value that researchers conducting a Northwestern University library experiment in 2005 sought to test in relation to the material in ECCO. In order to do this they gathered a sample of around 52,000 records from WorldCat, in which they found ‘a total of 30,948 different subject headings were applied 107,477 times – an average of slightly more than two subject headings per title’; these headings were then imported to their local catalogue records for ECCO (Reference GarrettGarrett, 2007: 71). The project tested the effectiveness of subject headings by pitting searches in the subject heading field against searches conducted just by title or just in the full text. In both cases the project found that searches using subject headings found significantly more texts than either title or full-text searches (Reference GarrettGarrett, 2007: 73–5). For example, a title may not contain the phrase you’re looking for, even though the book might; however, this does not necessarily mean that the full text of the book contains this phrase either: in the example just given, a search for ‘climatology’ would return no results at all since the term didn’t exist in the period. As Jeffrey Garrett puts it, subject headings are a ‘Meta-vocabulary’ and their utility in a full-text database like ECCO is how they enable ‘searching and discovery, across centuries and across languages’ (Reference GarrettGarrett, 2007: 75).

As we’ve seen, the search engine in keyword mode searches a number of metadata fields including the subject headings. But I want to dig a little deeper in order to differentiate this type of data from other ECCO fields, since this is not a purely machine-driven process: unlike the OCR text, or the title or author, the subject heading was assigned by a combination of human interpretation and computer-aided harvesting of data. Between 2005 and 2007 the Center for Bibliographic Studies and Research at the University of California, Riverside – the headquarters of the ESTC – began a project to update its subject headings. This was largely due to the fact that while records for material before 1700 incorporated subject headings, material after that date did not – in other words, the books catalogued by the 18thC STC and eventually digitised in ECCO (Reference GarrettGarrett, 2007: 70). Part of this project involved the Northwestern University experiment in 2005, but the ESTC also tasked HTC Global in 2007 to add subject headings to the ESTC’s catalogue records, starting with around 63,000 ECCO titles. In this project HTC operators examined the existing ESTC records, used the digitised images to read portions of the works, and finally assigned subject headings to each work using the Library of Congress subject headings list (Reference Geiger and SchillingGeiger & Schilling, 2016). Eighteenth Century Collections Online’s enhancement of its metadata and the MARC library records it sells in addition to the database was completed around 2009. In 2012 the interface’s ‘FAQs’ described how ‘[o]ver 274,000 subject headings’ were added ‘through the combination of harvesting and manual assignment’. Some subject headings were obtained from ‘existing’ library records which held the physical copy, and where this was not possible, ‘ESTC licensed the work of adding LoC headings’ (in other words, the HTC Global project).Footnote ⁴⁴

This suggests considerable potential for divergence between these two systems of gathering and assigning subject headings, driven as they were by different organisations and groups of people. In short, the subject headings can be idiosyncratic. I am reminded of a search I conducted when I wanted to find out what other works of fiction were published in the same year as Samuel Richardson’s Clarissa (1748) – a book I was teaching at the time. Using ‘Fiction’ and ‘English fiction’ as subject heading terms in ECCO I found twenty-two titles. However, the same search in the ESTC produced a different set of results. This rough-and-ready exercise set me on a different path and made me think about how these databases categorise print history (Reference GreggGregg, 2016). There are disparities in the bibliographical metadata between the ESTC and ECCO (Reference MayMay, 2009). But a few brief examples concerning the subject headings reveal much about the odd processes behind the creation and management of metadata in ECCO. Whoever tagged Ovid’s Heroides. English Ovid’s epistles … Translated into English verse as ‘fiction’ clearly didn’t understand the title. And why wasn’t Henry Fielding’s canonical novel The History of the Adventures of Joseph Andrews tagged as ‘fiction’ in ECCO? It wasn’t a case of looking at the title and thinking, ‘well, this is a history or a biography’. Both the ESTC and ECCO give this the specific subject heading ‘Tobacco-fiction’, a peculiarly specific heading assigned to just three titles in the entire ESTC; the other two are novels by Tobias Smollett: The Adventures of Peregrine Pickle (1751) and The Expedition of Humphry Clinker (1773). Now, it’s true that people smoke in these novels, but plenty of other protagonists from the fiction of the period smoke. Such idiosyncrasies are illuminating because the creation of this metadata reveals the tangled history of the 18thC STC, the ESTC, and ECCO, and the peculiar mix of machines and human agency that underpins the search interface.

Nevertheless, by 2009 Gale’s upgrades to ECCO were considerable. Not only had it added the subject field to its metadata, it had upgraded its OCR to ABBYY, and it had published ECCO phase II of another 46,000 titles. In addition, Gale was listening to its library users and scholars: from 2007 the ECCO interface included a cross-search facility with EEBO so that, if the library had access to both databases, users could conduct searches within the ECCO interface to access the pre-1700 material digitised in EEBO (Reference RogersRogers, 2007). Another example was the issue of ESTC metadata. Scholars such as Eleanor Shevlin were discussing how databases like ECCO might enable a more interactive relationship to ESTC data and ‘provide opportunities to correct bibliographical inaccuracies’ (in Reference BattigelliBattigelli, 2010). From 2012 ECCO’s ‘Full Citation’ window included an email link entitled ‘Report an issue with this ECCO/ESTC record’.

It should also be clear that ECCO has never been a static entity. Scholars using ECCO clearly influenced the upgrades to the original interface in these years. But the design of ECCO was forcefully shaped by the landscape of academic electronic publishing in the mid- to late 1990s. Gale perceived a gap in existing academic and educational digital resources and capitalised on its existing content: the microfilm collection The Eighteenth Century. To a significant extent, this constrained ECCO’s capabilities in terms of how the books looked and the effectiveness of full-text searching. Nevertheless, Gale deployed state-of-the-art digitisation and data handling technologies in order to push to the limits what could be achieved with microfilm. The sum of these decisions shaped how ECCO could be searched, how these books could be discovered, and how we users apprehend eighteenth-century books. The next chapter explores how concerns about the usability of the underlying OCR text, as well as the costs of access, led to a number of licensing deals and collaborative projects that – together with Gale’s own developments – offered new ways of using, accessing, and interfacing with ECCO.

Partnerships, Licensing, and Access, 2004–2020

The interface is not a transparent portal to access, as Tim Sherratt has argued. ‘Online interfaces make assumptions about the needs and desires of users. They do not merely provide access; they construct it by defining the types of interactions we can have with collections’ (Reference Sherratt, Lewi, Smith, vom Lehn and CookeSherratt, 2019: 4). The issue of ECCO’s OCR overlapped with a wider discussion about what ‘access’ meant in relation to how digital texts might be used and transformed outside the confines of commercial interfaces. At the same time, academics and libraries were also expressing concerns about the cost of accessing ECCO and what was becoming known as a ‘digital divide’ in higher education institutions. In effect, this continues our exploration of the cultural and technological issues surrounding the questions of how we access old books that started with our discussion of Robert Binkley and Eugene Power’s hopes for microfilm in the 1930s.

Eighteenth Century Collections Online’s first partnership was with the TCP. The TCP was directly responding to the question of how access to digital texts could be enhanced and transformed. It aimed to enable its partners of libraries and commercial publishers to

• pool their resources in order to create full-text resources few could afford individually

• create texts to a common standard suitable for search, display, navigation, and reuse

• collaborate with commercial providers, rather than constantly bargaining and competing with them.Footnote ⁴⁵

As we’ve seen, the project initially aimed to give users the ability to conduct full-text searches of the material in EEBO. Given the limitations of OCR software working on hand-press printed material, the texts selected by the TCP were transcribed by hand, and were to be ‘transcribed by more than one person, and the results are compared against one another, to generate electronic text that is 99.95 per cent accurate’.Footnote ⁴⁶ By 2019 EEBO-TCP had transcribed around 40,000 texts.Footnote ⁴⁷ The ECCO-TCP project began in 2004. However, it turned out to be on a much smaller scale than the EEBO-TCP project. This was arguably because its aims were less coherent than for EEBO-TCP, given that a searchable text was already available via ECCO. Transcriptions continued until 2011, although a requested selection of medical texts was transcribed in 2012 (Schaffner, 2019). The ECCO-TCP project was expected to transcribe around 10,000 texts, but only 2,473 texts were eventually created (Reference BlaneyBlaney, 2019).Footnote ⁴⁸ The texts were released to Gale in several batches between 2005 and 2012; in 2011, by agreement with Gale, 2,231 texts were released to the public, and were also searchable via 18thConnect (Reference MandellMandell, 2011; Schaffner, 2019).Footnote ⁴⁹ They were made available via Jisc’s Historical Texts platform in 2016.Footnote ⁵⁰ These were small numbers compared with EEBO-TCP. Funding was the problem: as early as 2006 the executive board was proposing ways to ensure future funding in order to ameliorate predicted budget deficits.Footnote ⁵¹ As the director of the TCP, Paul Schaffner, recalls, ‘we never received the financial support that we hoped for’, so the period between 2009 and the final transcription in 2012 saw the project gradually wind down (Schaffner, 2019). Shawn Martin (the TCP project librarian at Michigan and an executive board member) pointed out that while ‘one could argue that because of OCR text running behind the images in … ECCO, a TCP project is not needed, it is also true that many kinds of research need the added value that TCP offers to encourage electronic scholarship’ (Reference Martin and DeyrupMartin, 2009: 6). The 2,000-plus texts of ECCO-TCP are publicly available and are free to use, analyse, edit, modify, and even republish without copyright restrictions, and can be accessed via a number of digital archives and spin-off projects (Gregg, 2019). The ECCO-TCP project provided the ‘ground truth’ for the eMOP project to test the accuracy of its OCR engine’s performance against the hand-transcribed texts of the TCP (Reference MandellMandell, 2015: Appendix C). The TCP set also plays a part in a variety of archives and large-scale text analyses, such as the Eighteenth-Century Poetry Archive (2012), Visualizing English Print (2015 ongoing), Linguistic DNA (2015 ongoing), and Commonplace Cultures (2016 ongoing).

One of the TCP’s aims was to create access to texts ‘few could afford’. The crucial context in the late twentieth century that was shaping the debates about online access to scholarship was the drive towards ‘open access’. As Kathleen Fitzpatrick summarised it, this is the ‘ethical desire to break down the barrier between the information “haves” and “have nots” of the twenty-first-century university structure’ (Reference FitzpatrickFitzpatrick, 2011: 160). The notion of universal access to knowledge had been present since the early twentieth century, with thinkers and scholars such as Robert Binkley and H. G. Wells’ notions in his book World Brain (1938), but the advent of the World Wide Web and large digitisation projects accelerated the means and intensified the stakes of what ‘access’ meant. Martin Paul Eve identifies two socio-historical drivers: the ‘serials crisis’ in academic publishing and libraries, and the ‘free culture’ ethos (Reference EveEve, 2014: 12–21). What Eve terms the ‘serials crisis’ was the symptom of both an exponential increase in scholarly research in the last half of the twentieth century and an increase in publishers’ pricing of journal subscriptions that outstripped inflation and library budgets, resulting in ‘paywalls that hinder [the] ability to conduct research and to teach/learn efficiently’ (Reference EveEve, 2014: 13). Simultaneously, late twentieth-century commercialised information technology gave rise to a ‘counter-discourse of “free culture”’ which was articulated not solely in terms of cost, but more as the ‘freedom to reuse material’ (17). This addressed, in other words, the difficult relationship between the holder of the copyright to a work and the user’s ability – or not – to reuse that material. The two concerns about scholarly publishing and free culture eventually came together in a series of defining documents in 2002 and 2003 (21). We’ll see how this democratising impulse fed through to ECCO’s licensing deals and partnerships.

The debates around digitisation and access to knowledge were to become particularly febrile following the legal challenges to the mass digitisation of millions of books by Google Books. Its explosive programme of scanning in the early 2000s broached the border between out-of-copyright works and works still under copyright: its vision of universal access came forcefully up against the commercial interests of publishers and the rights of authors, resulting in a complex series of legal challenges and appeals by the Authors Guild of America (working with the Association of American Publishers) and Google between 2005 and 2014. In 2008, at the height of these proceedings, Google proposed to award a $125 million settlement in return for considerable rights to the material it had digitised. However, scholars, librarians, and internet activists were concerned that Google, far from freeing the world’s knowledge, was effectively monopolising it (Reference DarntonDarnton, 2012; Reference SomersSomers, 2017; Reference ThylstrupThylstrup, 2018: 45–8).Footnote ⁵²

These arguments were echoed in 2015, albeit briefly, between scholars and the publisher of EEBO, ProQuest. Unusually, since EEBO was a library-only subscription, the very large membership of the Renaissance Society of America had been granted access; however, in October 2015 ProQuest cancelled that subscription because, as the society surmised, its members’ ‘heavy use’ threatened ProQuest’s ‘potential revenue from library-based subscriptions’ (cited in Reference WexlerWexler, 2015). The reaction from the society’s membership was immediate, as users took to Twitter to complain with the hashtags #EEBOgate, #ProQuestGate, and #FrEEBO. ProQuest quickly reversed its decision, but users were outraged by how easily commercial publishers could bar access to research. John Overholt thought this should be a ‘wakeup call … These books are part of our cultural heritage, and it’s high time we made them available to everyone’ (Reference OverholtOverholt, 2015). #EEBOgate certainly gained traction coming so soon after the collapse of the case against Google Books in 2014.

This discourse about paywalls and free culture was also heard in earlier questions about Gale’s ECCO. Certainly, it’s a fact that ECCO’s price point kept it out of the smaller colleges and universities. ‘Access doesn’t come cheap’, one reviewer commented at its launch, while another noted that its ‘price tag may be a hurdle for many libraries’ (Reference SmithSmith, 2003; Reference LevackLevack, 2003). In December 2009 Peter Reill (then president of the American Society for Eighteenth-Century Studies), sent a message via the email discussion group C-18 L to address ‘the question of the increasingly unequal access of scholars to digital resource databases that are critical to pursuing research in their fields’, and set out a number of questions for the society to debate:

The issue of affordability for libraries mentioned in some of the first reviews of ECCO had by 2009 been transformed into the issue of ‘unequal access’. The issue was picked up by the Early Modern Online Bibliography blog, which generated a number of comments; in one, Anna Battigelli argued that:

As Battigelli draws out the interdependency of library resources, staffing, and student recruitment, she ends with the striking conclusion: ‘The health of eighteenth-century studies – or of any form of early modern studies – is at stake’ (in Reference ShevlinShevlin, 2009).

One unique response to the digital divide was Ben Pauley’s ‘Eighteenth-Century Book Tracker’, an online archive created as a solution to the lack of access at his own college. As Pauley describes it, this was intended to be ‘a clearinghouse for registering links to freely-available digital facsimiles of eighteenth-century texts, pooling its users’ discoveries and attaching them to bibliographically responsible entries’ (Reference PauleyPauley, 2009). Book Tracker was an attempt to offer an alternative to Gale’s pay-walled ECCO by collating and indexing links to eighteenth-century books digitised elsewhere (largely, Google Books) and contributed by a scholarly community of users. However, from 2019 the site lost its crowdsourcing functionality: it was, as its creator acknowledged, a one-person operation and difficult to sustain.Footnote ⁵³ Pauley’s project brings to the fore the tension within the wider debate about open access: the long-held idealism of universal knowledge and the economic realities required to enable and sustain such ideals.

There is actually a non-digital way of accessing ECCO titles. From 2010 BiblioLife began producing print-on-demand (POD) copies of Gale’s digital collections under the licensed ‘ECCO Print Editions’ imprint. This comprised around 117,000 titles; as with ECCO, Gale pays royalties to the source library for every copy sold (Dawson, in Reference ShevlinShevlin 2010). In POD publishing the electronic files are held by the publisher and copies are printed and bound only when a customer, via an online retailer such as Amazon, orders a copy. Print-on-demand publishing owed much to the availability of digital printers in the 1990s that could print off small numbers at low unit cost, and was initially used to print technical instruction manuals (Reference ThompsonThompson, 2005: 421–2). Later the giant Ingram publishing company saw the possibilities for revolutionising the business of publishing books; as one source put it, ‘It used to be print book, sell book. We say, no, no. Sell book, print book’ (quoted in Reference ThompsonThompson, 2005: 422).

Indeed, BiblioLife does not define itself as a publisher: ‘We are really a software company that has books coming out at the end of our process’ (quoted in Reference AlbaneseAlbanese, 2010). BiblioLife is based in Charleston, South Carolina, and is the umbrella company to a number of POD subsidiaries including BiblioBazaar and Nabu Press, and it is owned by BiblioLabs LLC (Wikipedia Contributors, 2019). BiblioLife specialises in publishing historical reprints, and its president defined its business model as ‘really focused on unique materials that are not part of mass digitization projects … Who has that content and how we are getting it is something that is a competitive advantage’ (quoted in Reference AlbaneseAlbanese, 2010). However, this pugnacious self-image of BiblioLife is slightly contradicted by its more utopian blurb inside the covers of ECCO Print Editions. In this POD copy of a 1704 pamphlet by Daniel Defoe BiblioLife explains its part in a project called the BiblioLife Network:

Rather than leveraging a niche marketing opportunity, BiblioLife’s mission is ostensibly part of a long-standing ethos to preserve and provide access to important historical heritage – which is perhaps beside the point, because in one way, by making available low-demand historical materials at low cost (relative to institutional digital access costs), there is an echo of Binkley’s and Power’s hopes for accessing old books I explored earlier in relation to microfilm.

BiblioLife’s branding – ‘old books, new life’ – might alert us to consider what kind of ‘life’ this is. The reaction to some POD copies of old books homed in on their peculiar lack of bookishness: many do not include information about how they are produced, or what edition or source copy has been used (Reference ShevlinShevlin, 2010). It is for this reason Whitney Trettien termed these books ‘zombie-like’, as if their previous life has been cut away (Reference TrettienTrettien, 2013). However, BiblioLife’s ECCO Print Editions are unusually forthcoming on metadata and the processes of remediation. ECCO’s metadata and ESTC numbers are reproduced in Amazon’s product description and in the inside cover of each POD book ‘in the interest of increasing the chance of edition identification’ (Scott Dawson, in Reference ShevlinShevlin, 2010).

Gale’s own inside cover blurb also emphasises the ability to access its digitised books outside the paywall of a subscription database: ‘Now for the first time, these high-quality digital scans of original works are available via print-on-demand, making them readily accessible to libraries, students, independent scholars, and readers of all ages’ (Reference DefoeDefoe, 2010). The argument was echoed by scholars and couched in terms of widening access: ‘One thing that is great about ECCO’s POD service through BiblioLife is that those without access to ECCO can get access to selected works, admittedly at a price. This may not be the perfect solution to access problems, but it is a bit better than having no access at all’ (Battigelli, in Reference ShevlinShevlin, 2010).

Other scholars and institutions were collaborating with Gale to ameliorate the digital divide and to widen access. One is the US-based academic community platform 18thConnect. This was linked to eMOP, the OCR project already mentioned, and was initially conceived in 2009 at a digital humanities meeting in Dublin.Footnote ⁵⁴ 18thConnect acts as an open-access searchable aggregator platform for a large number of eighteenth-century resources, including ECCO. Users find results in ECCO but still need institutional access to click through to the images in the collection. Its founder, Laura Mandell, had also wanted to create a tool to correct ECCO’s OCR. In 2010 Gale granted 18thConnect ‘a limited, nonexclusive, royalty-free right to use the Typed Plain Text, OCR Plain Text and Metadata’ (Reference MandellMandell: 2012b: 23). In July 2010 the National Endowment for the Humanities granted an award of $41,000 for ‘18thConnect: an open access resource’, out of which ‘Typewright’, a crowdsourcing correction tool, was developed (Reference MandellMandell, 2019a). The Typewright interface sits on the 18thConnect website and enables users to pull up the raw OCR text from ECCO underneath a window which shows a snippet view from the corresponding page image; users then manually correct the OCR text. However, once a complete text has been corrected, 18thConnect sends the user/editor the full corrected text ‘to use as he or she likes’, but principally with the hope of creating scholarly editions, echoing the aims of open access and scholarly reuse exemplified in the TCP project (Reference MandellMandell, 2012a: 302; emphasis in the original).Footnote ⁵⁵

The most significant alternative access to ECCO – for the United Kingdom – is Jisc’s platform, Historical Texts. The Joint Information Systems Committee (Jisc) was formed in 1993 in order to manage and develop UK network and information systems for an ever-expanding higher education system.Footnote ⁵⁶ After the 1993 Follet report on university libraries, Jisc was allocated £15 million to create and oversee various projects that would enhance its digital capabilities, called the ‘Electronic Libraries Programme’ (‘eLib’).Footnote ⁵⁷ Jisc began to function as national negotiator enabling access to licensed content on behalf of the UK education community, which enabled Jisc to achieve significant cost benefits. Notably, Jisc also enables UK Further Education institutes, with students aged between sixteen and eighteen, to access digital resources.

Jisc Collections’ major coup was signing licensing agreements with ProQuest for EEBO, and with Gale for ECCO I in 2006 and for ECCO II in 2009.Footnote ⁵⁸ These would form the basis for Jisc’s Historic Books platform, which began development in 2011, and was redesigned in 2014 as Historical Texts. The platform now offers access to ‘over 460,000 late 15th to 19th century texts’, including the British Library’s 19thC Collection and non-paywalled access to the UK Medical Heritage Library, and there are plans to add access to the Burney Collection of eighteenth-century newspapers.Footnote ⁵⁹ The deal included page image files and the XML files for the OCR text plus bibliographical metadata; they were bought from Gale in perpetuity, while Gale retained the copyright for the images (Reference GibbensGibbens, 2019). When it came to the MARC library records for the material (these enable each individual record to be loaded on to a library’s catalogue) Jisc bypassed the necessity for buying MARC records from Gale by securing a partnership deal directly with the ESTC and then added its own enhancements, thereby securing a better deal for UK higher education institutions. In addition, Historic Texts was able to ingest the ECCO-TCP set (the ECCO texts transcribed by the TCP), since Jisc had been a member of the partnership (Reference Marchionni and MilloyMarchionni & Milloy, 2020).

Historical Texts is part of the ‘Jisc eCollections’ programme’ launched in 2011. As Caren Milloy (the director of licensing since 2020) described it, this is ‘a community-owned content service developed to protect and preserve existing content investments by offering an alternative to commercial providers. It brings access fees within the control of the community, in terms of ring-fenced reinvestment and moderation of increases’ (Reference MilloyMilloy, 2012). A particular source of frustration in UK libraries is the pricing structure of digital collections. Jisc’s reports ‘Digital Archive Collections Platform Charging Survey’ (2018) and ‘Library Collections: Navigating the Payment for Access Minefield’ (Reference FindlayFindlay, 2019) highlight how, on top of the one-off perpetual cost, additional costs are associated with yearly ‘hosting’ or ‘platform’ fees, and there is also the cost of hard drives publishers sell in order to enable users to conduct text analysis (known as ‘TDM’ drives). Gale’s TDM drives, which contain OCR text and metadata, cost between $500 and $1000 per resource, more if it is a particularly large collection (Gale, 2014). On hosting or platform fees, one report found that in the ‘last academic year’ 46 per cent of respondents spent up to £5,000, while 33 per cent spent more than £15,000 on these fees (Jisc, 2018, in Reference FindlayFindlay, 2019).

It is tempting to see Jisc’s Historical Texts as a competitor of Gale’s ECCO in the United Kingdom, but this is to ask the wrong question. In an interview with Paola Marchionni (the head of digital resources for teaching, learning, and research) she acknowledged that Jisc wants to provide a ‘better deal’ for the UK education community and to recoup costs. However, she insisted that ‘we don’t see ourselves as publishers … the drivers are different.’ Marchionni contrasted the effectiveness of Jisc to enable access to digital content in the United Kingdom with the bigger digital ‘divide’ in academic institutions in the United States and Australia, and she was emphatic that Jisc’s aim is to ‘democratise access across the sector’ (Reference Marchionni and MilloyMarchionni & Milloy, 2020).

Academics were aware of the economics of sustaining digitisation, but there were tensions in attitudes. Laura Mandell recalled her initial outrage at the commercialisation of public heritage, but while developing the eMOP project at Texas University, she acknowledged that collaborative projects between libraries, academics, and commercial publishers like Gale are capable of effectively sustaining scholarly resources (Reference MandellMandell, 2015: 2–3). In relation to the 18thConnect platform, Mandell noted her ‘thanks to Gale for its openness to scholarly needs’.Footnote ⁶⁰ However, community, openness, and commercial profit are difficult philosophies to reconcile. More recently, Gale’s presence at conferences, both as keynote speaker and as sponsor of those conferences, has caused some concern. During my own talk about ECCO at the Digital Humanities Congress in Sheffield in 2018, I was bemused at my position: accepting a glass of Gale’s sponsored wine just after having offered a critique of ECCO. Concerns were more seriously voiced during the large annual international digital humanities conference organised by the Alliance of Digital Humanities Organisations in Krakow in 2016. Conference goers took to Twitter to express their discomfort: for many the issue was that Gale’s presentation – about its TDM drives – sounded more like a promotion of its product and so directly opposed the spirit of scholarship. In a follow-up piece, ‘What Price Gale Cengage?’, Andrew Prescott summarised these tensions nicely: while acknowledging that sponsorship is valuable for the running of large conferences, he noted that the ‘production by Gale of enormously expensive digital packages which can only be afforded by university libraries seems at odds with the open access aspirations of many digital humanities practitioners’. The desire that ‘contents of libraries, archives and museums owned by the public and part of a shared cultural heritage’ should not be behind an expensive paywall reflected the aspirations of those who viewed with dismay #EEBOgate and the Google settlement discussions (Reference PrescottPrescott, 2016).

Platforms

On one hand, Gale’s genuine wish to engage with its users echoes the early aims Macomber voiced when ECCO was launched in 2003 (Reference HaneHane, 2004). On the other hand, Gale is a commercial entity that has to sustain its proprietary investments and to recoup costs through development and sales. Eighteenth Century Collections Online’s development in the 2010s was influenced by the techno-commercial move towards platforms and apps, and also by its increasing awareness of digital humanities scholarship and practice, and in particular the use of computer-aided text analysis. Its response leveraged the idea of an enhanced interface that would enable users to directly work with text from across a diverse range of its products, resulting in the Gale Primary Sources and Gale Digital Scholar Lab platforms, launched in 2016 and 2019, respectively. Accessing ECCO via either Gale Primary Sources or Gale’s Lab, or even Jisc’s Historical Texts, is a fundamental shift away from the conception of an interface as the identifying face of a particular online collection. In 2020 Gale staff called the original ECCO interface the ‘legacy’ and ‘stand-alone’ interface, terms that together suggest that, as an interface designed solely for the digitisation of as single collection, it is outdated and it is time for it to be superseded (Reference SullivanSullivan, 2020). As I write, ProQuest is also scheduled to retire its stand-alone interface for EEBO and replace it with the ‘ProQuest Platform’.Footnote ⁶¹ Welcome to the age of the platform.

The ‘rise of the platform’ – or ‘platformisation’ – has been associated with the big social media products such as Facebook, Twitter, and YouTube, but also with mass digitisation projects such as Google’s, and even with how libraries can be conceptualised (Reference WeinbergerWeinberger, 2012; Reference MatternMattern, 2014; Reference HelmondHelmond, 2015; Reference ThylstrupThylstrup, 2018: 126–31). But this use is itself built on older meanings. Tarleton Gillespie, developing the definitions offered by the OED, identified ‘four semantic territories that the word “platform” has signified in the past’ (Reference GillespieGillespie, 2010: 349). The first of these meanings is computational: ‘an infrastructure that supports the design and use of particular applications’ such as operating systems, hardware, or online applications and tools (Reference GillespieGillespie, 2010: 349). Gale uses the term in this sense to describe its internal digital infrastructure. Eighteenth Century Collections Online is separate from its interfaces, and is a collection amongst many within Gale’s internal platforms. For instance, the original stand-alone interface for ECCO is underpinned by a backend platform linked to a search engine it shares with other collections (such as British Library Newspapers and the Burney Collection). These collections are scheduled in 2020–1 to be migrated to a new backend platform driven by a content repository and an updated and highly modified version of the Lucene search engine which, in turn, underpins the Gale Primary Sources and Gale Digital Scholar Lab platforms (Reference SullivanSullivan, 2020).

This is of critical consequence for search results: they will be different depending on which interface or platform you are using (or have used). For simplicity’s sake we can test this using the singe-box ‘basic search’ function and try the word ‘dog’. This produced the following hits:

• Original, stand-alone interface: 52,449

• Gale Primary Sources: 73,686

• Gale Digital Scholar Lab: 73,686

• Historical Texts: 58,848

The last set of results illuminates what Kelly Centrelli noted in an early review of Jisc’s Historic Books (the first version of Historical Texts): that the same search in ECCO and in Historic Books produced ‘completely different’ results because of its ‘contextual word’ searches (Reference CentrelliCentrelli, 2012). The results are produced by the configuration of Jisc’s own backend platform, its own search engine, and a particular search algorithm: the history of this particular platform also reveals a history of changing backend systems.Footnote ⁶² In short, search results are shaped by the search algorithms peculiar to each interface.

Each of Gale’s interfaces uses a customised search algorithm, or what Gale terms a ‘search recipe’. I will exemplify how such an algorithm works for the ‘basic search’ function in Gale’s Lab in order to explore how its search recipe builds and sorts results according to ‘relevance’. The search recipe ‘specifies which indexes to search and the proximity operators to apply. This determines how many articles are retrieved, and what the relevance score of each of those articles will be’ (Reference SullivanSullivan, 2020). The ‘basic search’ recipe looks like this:

{stemming=true;subjectExpansion=true;fields=TI|SU|KE|TXT;
operator=N4} (Houghton, 2020)

Gillespie outlines other meanings of the platform; the oldest of which is architectural, as in a physical structure upon which people or things stand. The further, figurative meaning extends this sense of an architecture as a metaphor ‘for opportunity, action and insight’ (Reference GillespieGillespie, 2010: 350). Jisc’s Historical Texts echoes the architectural sense of the term, describing itself as a ‘platform’ that ‘brings together four historically significant collections for the first time’.Footnote ⁶⁴ Other digital collections mobilise both figurative and computational meanings. The Digital Public Library of America (DPLA), for example, launched itself as both a ‘portal’ and a ‘platform’; exploring the DPLA, but also digital libraries such as Europeana and Trove, Tim Sherratt usefully articulates the difference: ‘Portals are for visiting, platforms are for building on’ (Reference SherrattSherratt, 2013). These digital collections, like Hathi Trust Digital Library or Old Bailey Online, all provide application programming interfaces (APIs) that allow users to bypass the search interface and work directly with the archive’s text or underlying metadata, or using third-party tools that are linked via the collection’s website to analyse or transform that material. Both Gale Primary Sources and Gale Digital Scholar Lab exemplify the figurative and computational senses of the term: they are platforms in the sense of an architecture upon which stand a large number of Gale’s digital collections. Gale Primary Sources was described as an ‘Enhanced product platform’, but they both also offer applications to analyse the texts; the Lab offers tools and third-party applications and is described as a ‘text and data mining platform’ (Reference Houghton and KetchleyHoughton & Ketchley, 2019: 5; Reference Rand and FustRand & Fust, 2019).

This is perhaps the most radical departure from bookishness for ECCO’s books since both platforms were explicitly designed to enhance the analysis of the OCR text in the light of digital humanities and computer-aided quantitative analysis, popularly known as ‘distant reading’ after Franco Moretti’s work (also known as text analysis, text mining, or macroanalysis) (Gale, 2016, 2018). The earliest use of computers in humanities scholarship was for linguistic analysis and electronic concordances, such as Father Roberto Busa’s computer index to Thomas Aquinas’ Summa Theologiae, begun in 1949 with IBM. Busa is generally cited as the seminal figure in humanities computing (Reference WisbeyWisbey, 1962: 163–7; Reference Hockey, Schreibman, Siemans and UnsworthHockey, 2004: 3–19). However, the increasing availability of very large collections and archives of digitised text by the early twenty-first century meant that the ‘number-crunching abilities of the computer’ could be ‘set free to perform quantitative analyses that shed light on the text’s authorship, style, genre, theme, plot, even ideology’ on an ‘unprecedented scale’ (Reference HammondHammond, 2016: 83; Reference JockersJockers, 2013, 27).

In 2013–14 Gale launched two initiatives to make its digitised collections more amenable to such techniques and tools. One was by selling TDMs to subscribing libraries, essentially a terabyte hard drive that contained the XML files of the metadata and the OCR text (Gale, 2014). The other was the roll-out of a new interface to some of Gale’s collections, including ECCO, called ‘Artemis’, which allowed the user to directly see and work with the OCR text using built-in tools for word frequency analysis and visualisation (Gale, 2013). Artemis was short-lived and was rapidly replaced in 2016 by Gale Primary Sources. While we’ve seen how the TDM drives have been problematic for libraries, the initiative was Gale’s response to digital scholarship. Chris Houghton (Gale’s head of digital scholarship,), discussed in an email interview what led to the development of the Gale Digital Scholar Lab:

One of the results, he notes, was that ‘we made the conscious decision to be brave and really lift the lid on Gale’s OCR process’ on the new interfaces. The perception of the open culture of digital humanities was perceived to go hand in hand with scholars’ need to work directly with the OCR text and metadata of Gale’s digital collections: ‘We knew that the community prized openness and accessibility, which would automatically clash with the fact that our archives are behind paywalls, so we came to the conclusion that we would try and make as much of the platform as open and visible as possible, allowing users to extract data and analyses at every stage’ (Reference HoughtonHoughton, 2019). Gale’s hope was that the sophisticated capabilities of its latest interface to its digital collections like ECCO would somehow mitigate or offset the effect of paywalled access.

Gale has certainly worked at being more transparent about what is involved in digitisation. On the Gale Primary Sources public page there is a now a ‘Behind the Scenes’ page with an article entitled ‘Creating a Digital Archive: The Technical Processes’ which takes us through the basics aspects of image capture, OCR, metadata, XML, quality assurance, and last ‘the application’ (in other words the interface).Footnote ⁶⁵ In parallel, both of the latest interfaces to Gale’s digital archives offer the ability to open a side window to see the OCR’d text accompanying the page image.Footnote ⁶⁶ Both also offer a screen-tip that explains ‘How this text was created’ which notes factors that affect OCR confidence, including the ‘condition of the original document’, whether it is handwritten or printed, the date the document was created (this seems to mean when the remediated document was created), the scanning equipment, and the ‘Maturity of the OCR algorithm used at the time of creation’. Moreover, each document’s text comes with a figure that represents the OCR engine’s ‘confidence in the accuracy of the conversion from image to text’. The visibility of the OCR confidence figure is potentially useful. One study has found that, while ECCO’s figures are on the whole higher than other calculations, there was a general correlation between Gale’s figures and those of the study, indicating that, relatively speaking, this ‘supports being able to use the OCR engine confidence value to accurately assess OCR quality’ (Reference Tolonen, Mäkelä, Ijaz and LahtiTolonen, Mäkelä, Ijaz, & Lahti, forthcoming 2021). However, ECCO II OCR confidence figures are lower than for ECCO I (Reference Tolonen, Mäkelä, Ijaz and LahtiTolonen et al., forthcoming 2021). This is probably because they are calculated in different ways by the two different OCR engines: PrimeRecognition was used on ECCO I in 2003; ABBYY was used for ECCO II in 2009.Footnote ⁶⁷

The text analysis tools on the Lab platform are powerful and capable of tracing highly suggestive topical and linguistic patterns. The platform offers tools for cleaning the OCR text, word frequency counting (ngrams), document clustering, topic modelling, sentiment analysis, and natural language processing. In addition, all the tools are capable of producing data in the form of spreadsheets or various kinds of graphic visualisation. However, the increased focus on the text of the books in ECCO risks bypassing the bookishness of books, those eccentricities introduced by the handmade processes of book production and transmission. This shift can be traced in the presentation of metadata on these two platforms.

In Gale Primary Sources the reader can choose to prioritise page images or text, or have the page image surrounded by text, a table of contents, an ‘explore’ window (a facility to search the text of the book), or bibliographical metadata (‘full citation’). Curiously, metadata is split between two different views. The first, in the ‘explore’ view, contains a link that opens up a pop-up window listing the holding libraries, a brief note about pagination, and what to many must be an obscure reference, ‘Moore, 68’ (Figure 22).Footnote ⁶⁸ There is also an option to see the Essay alongside an expanded set of metadata (Figure 23).

Figure 22 Gale Primary Sources, ‘Explore’ view

(screenshot, 9 July 2020)

Figure 23 Gale Primary Sources, ‘Full Citation’ view

(screenshot, 9 July 2020)

It is not clear why different facets of the book’s metadata are split across separate windows. Neither is it clear for whom this selection of metadata is intended: the Gale ID seems to look towards internal systems; some of the bibliographical metadata is very specialised (e.g. ‘Moore, 68’, or the microfilm reel number), while some is suppressed (no LoC subject headings); and the document type ‘monograph’ might be of utility to a library cataloguer but likely not to a scholar or student.

The way in which Gale Digital Scholar Lab presents a book’s metadata is again different. It includes only basic metadata of author, title, and imprint, Gale’s document identification, a link path to ECCO, and the date of access. The source library is indicated in the results page, but all other material traces of the book – the bibliographical metadata accumulated by the ESTC – have not been included. This means that, for example, if we wanted to search for the particular edition of Alexander Pope’s Works (1736), for which I examined the title page to volume one in my earlier section on bookishness, we can see that ECCO digitised three copies. Each has small material differences; the ESTC records these differences and allocates them individual ESTC identification numbers. However, since the Lab platform does not reproduce this, it’s very difficult to tell which particular copy is which (Figure 24).

Figure 24 Gale Digital Scholar Lab, search results list

(screenshot, 1 June 2020)

Bibliographical metadata is also there to help us decode or add material context to what cannot be seen in the image. In the example of the title page to volume one of the Works, there is no bibliographical metadata, no indication at all, that the title page is in fact printed in red and black ink (Figure 13). So, while clearly extending the capabilities of computing power to analyse text, the Lab has returned us to a state before ECCO where image and bibliographical metadata are separated. This also parallels some of the conceptions of distant reading, which, as Katherine Bode has pointed out, conceives text as the be-all and end-all, in which works are perceived as ‘singular and stable entities’ abstracted from the medium of their transmission – the book. Bode contrasts this perception with an understanding of the works as ‘events – unfolding and accruing meaning across time and space – and the documentary record as partially and provisionally expressing that process’ (Reference BodeBode, 2019: 5). In short, the platform risks the very abstraction that Robin Alston, the editor of the 18thC STC, worried about when negotiating between the organisation of data required by computerised cataloguing systems and the ‘eccentric evidence’ of books produced by hand-press (Reference AlstonAlston, 1981a: 379). Jisc’s Historical Texts, Gale Primary Sources, and Gale Digital Scholar Lab are designed to offer libraries access to cross-searchable packages of different digital collections, each of which were created differently and with different metadata. Standardising how metadata was represented in these platforms was an understandable and challenging priority for Jisc and Gale. In addition, the age of quantitative analysis has radically shifted the priorities of how a book is understood and interrogated. These factors are clearly reflected in the decreasing coherence and detail of bibliographical metadata and an increased focus on text in these platforms.

In addition, the subsuming of individual collections like ECCO within a single platform homogenises the nature of these heterogeneous collections. The history of each collection’s creation will be different and exerts a decisive – if sometimes obscure – force on the nature of the collections. In a recent study researchers compared the entire ECCO corpus to the most accurate data of the print culture of the period, the ESTC, in order to assess the issue of representativeness (ECCO includes around 50 per cent of what is in the ESTC). However, the study found a number of ‘clear biases in the collection’ that are a result of its ‘complicated provenance’ (Reference Tolonen, Mäkelä, Ijaz and LahtiTolonen et al., forthcoming 2021). For example, it notes that material printed in America is under-represented relative to material printed in England, Scotland, Ireland, France, and other countries, and perhaps unsurprisingly, works in the English language dominate both ECCO and the ESTC (although ECCO includes a large proportion of works in Welsh). Moreover, the distribution of texts from across the century is biased: ECCO contains far more works from 1780–90 and 1790–1800 than from earlier decades. Complicating this distribution is the presence of reprints of existing works, which represent a significant proportion of the total works in ECCO, and which vary from 20–25 per cent in the first decade of the century to 30–35 per cent between 1730 and 1790. Overall, these biases amplify each other and therefore crucially affect searches performed on ECCO’s interfaces (Reference Tolonen, Mäkelä, Ijaz and LahtiTolonen et al., forthcoming 2021).Footnote ⁶⁹

But the concerns of bibliography that are a result of ECCO’s ‘complicated provenance’ are also connected to the politics of representation, and we must attend to the cultural resonances of digital collections and their interfaces. Gillespie’s final meaning of ‘platform’ builds on the figurative and architectural: it is political, in the sense that a political organisation or politician might take a stand on or take up a position in relation to a certain issue (Reference GillespieGillespie, 2010: 350). This last meaning makes visible a tension within the idea of the platform: that it is at once a democratic, neutral, open space and a space that enables a particular position to be taken up, a space that is implicitly hierarchical or even exclusionary (Reference GillespieGillespie, 2010: 350–1; Reference ThylstrupThylstrup, 2018: 126–31).

Archives, as Trouillot remind us, are sites of power and also silence (Trouillot, 1995: 26). The urgent debates within the field of digital humanities have sought to reveal the silences of digital scholarship and how digitisation projects tended to replicate ‘a canon that skews toward traditional texts and excludes crucial work by women, people of color, and the GLBTQ community’ (Reference EarhartEarhart, 2012). As Tim Sherratt argues:

The prehistory of ECCO was, perhaps unsurprisingly, inflected by the cultural values present during the heyday of the massive cataloguing and microfilming projects of the twentieth century. Molly Hardy, reflecting on the records of African-American printers, reminds us that cataloguing is ‘built out of and therefore reflects on a particular moment’, and Tim Hitchcock highlights how the major commercial online collections are shaped by their dependence on microfilm collections whose selection criteria reflected the value placed on ‘male and European culture’ (Reference HardyHardy, 2016; Reference HitchcockHitchcock, 2016). The 18thC STC was an Anglo-American collaboration that projected a distinctly Anglocentric printed world. The decision for its geographical and linguistic focus was certainly driven by understandable, pragmatic reasons of scope and cost, but the 18thC STC as a whole was also shaped by the idea, as Alston put it, of a ‘national bibliographical record’ (quoted in Crump, in Reference Snyder, Snyder and SmithSnyder 2003: 49). Moreover, the initial phase of microfilming by RPI selected twenty-eight ‘major authors’ deemed worthy to have every variant and edition filmed, and who are therefore disproportionately represented in ECCO.Footnote ⁷⁰ The linguistic and geographical bias has its basis in the catalogue, but the effect of this is amplified by the initial emphasis on canonical authors – all male and all white – originating in the filming programme.

Nevertheless, access is not just about content, but also about what users can do with the content. As Roopika Risam argued, ‘[t]he reification of canons in digital form is not only a function of what is there – what gets digitalized and thus represented in the digital cultural record – but also how it is there’ (Reference RisamRisam, 2018: 17). For instance, the affordances of digitisation might offer ways of aiding the discoverability of voices obscured by the originary biases of digital collections by leveraging how texts and their metadata are organised, searched, and analysed (Reference KohKoh, 2014: 388; Reference RisamRisam, 2015). This arguably is what the platform might enable. The cross-searchable platform and the aggregation of different collections that include those with a wider linguistic and global scope is a potential means to ameliorate the biases of individual collections. In addition, while the Gale Primary Sources and Gale Digital Scholar Lab obscures the bookishness of old books, their focus on other forms of organising and discovering – in particular, the emphasis on text mining, visualisation, and machine-aided indexing – might be more sympathetic to the aims of postcolonial and intersectional digital humanities practices.

The platforms that give access to ECCO produce ambivalent effects. On one hand, they can leverage powerful tools for discoverability and, in some cases, analysis of hidden voices. To an extent this could mitigate biases hidden within the prehistory of individual collections of old books. However, the emphasis on text analysis comes at the cost of the material life of old books; subsumed within the platform, the material prehistory of collections like ECCO risks being lost.